diff --git "a/text/L1024/CuaS1Decoder_part3.mlmodelc/model.mil" "b/text/L1024/CuaS1Decoder_part3.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/text/L1024/CuaS1Decoder_part3.mlmodelc/model.mil" @@ -0,0 +1,7580 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] +{ + func main(tensor cos, tensor hidden, tensor last_onehot, tensor sin) { + int32 var_37 = const()[name = string("op_37"), val = int32(-2)]; + int32 var_58 = const()[name = string("op_58"), val = int32(-1)]; + fp16 var_57_promoted_to_fp16 = const()[name = string("op_57_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_67_cast_fp16 = pow(x = hidden, y = var_57_promoted_to_fp16)[name = string("op_67_cast_fp16")]; + tensor var_69_axes_0 = const()[name = string("op_69_axes_0"), val = tensor([-1])]; + bool var_69_keep_dims_0 = const()[name = string("op_69_keep_dims_0"), val = bool(true)]; + tensor var_69_cast_fp16 = reduce_mean(axes = var_69_axes_0, keep_dims = var_69_keep_dims_0, x = var_67_cast_fp16)[name = string("op_69_cast_fp16")]; + fp16 var_70_to_fp16 = const()[name = string("op_70_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_71_cast_fp16 = add(x = var_69_cast_fp16, y = var_70_to_fp16)[name = string("op_71_cast_fp16")]; + fp32 var_72_epsilon_0 = const()[name = string("op_72_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_72_cast_fp16 = rsqrt(epsilon = var_72_epsilon_0, x = var_71_cast_fp16)[name = string("op_72_cast_fp16")]; + tensor x_1_cast_fp16 = mul(x = hidden, y = var_72_cast_fp16)[name = string("x_1_cast_fp16")]; + tensor var_75_to_fp16 = const()[name = string("op_75_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor x_3_cast_fp16 = mul(x = x_1_cast_fp16, y = var_75_to_fp16)[name = string("x_3_cast_fp16")]; + tensor layers_0_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_0_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5248)))]; + tensor linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41948352)))]; + tensor linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_linear_attn_in_proj_qkv_weight_to_fp16, x = x_3_cast_fp16)[name = string("linear_0_cast_fp16")]; + tensor input_1_perm_0 = const()[name = string("input_1_perm_0"), val = tensor([0, 2, 1])]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("constant")]; + fp16 const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = fp16(0x0p+0)]; + tensor input_1_cast_fp16 = transpose(perm = input_1_perm_0, x = linear_0_cast_fp16)[name = string("transpose_55")]; + tensor input_3_cast_fp16 = pad(constant_val = const_1_to_fp16, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(8192)]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1])]; + tensor layers_0_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_0_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41964800)))]; + tensor input_5_cast_fp16 = conv(dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_linear_attn_conv1d_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + tensor var_104_cast_fp16 = silu(x = input_5_cast_fp16)[name = string("op_104_cast_fp16")]; + tensor qkv_1_perm_0 = const()[name = string("qkv_1_perm_0"), val = tensor([0, 2, 1])]; + tensor var_106_begin_0 = const()[name = string("op_106_begin_0"), val = tensor([0, 0, 0])]; + tensor var_106_end_0 = const()[name = string("op_106_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_106_end_mask_0 = const()[name = string("op_106_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_1_cast_fp16 = transpose(perm = qkv_1_perm_0, x = var_104_cast_fp16)[name = string("transpose_54")]; + tensor var_106_cast_fp16 = slice_by_index(begin = var_106_begin_0, end = var_106_end_0, end_mask = var_106_end_mask_0, x = qkv_1_cast_fp16)[name = string("op_106_cast_fp16")]; + tensor var_107 = const()[name = string("op_107"), val = tensor([1, 1024, 16, 128])]; + tensor q_1_cast_fp16 = reshape(shape = var_107, x = var_106_cast_fp16)[name = string("q_1_cast_fp16")]; + tensor var_109_begin_0 = const()[name = string("op_109_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_109_end_0 = const()[name = string("op_109_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_109_end_mask_0 = const()[name = string("op_109_end_mask_0"), val = tensor([true, true, false])]; + tensor var_109_cast_fp16 = slice_by_index(begin = var_109_begin_0, end = var_109_end_0, end_mask = var_109_end_mask_0, x = qkv_1_cast_fp16)[name = string("op_109_cast_fp16")]; + tensor var_110 = const()[name = string("op_110"), val = tensor([1, 1024, 16, 128])]; + tensor k_1_cast_fp16 = reshape(shape = var_110, x = var_109_cast_fp16)[name = string("k_1_cast_fp16")]; + tensor var_112_begin_0 = const()[name = string("op_112_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_112_end_0 = const()[name = string("op_112_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_112_end_mask_0 = const()[name = string("op_112_end_mask_0"), val = tensor([true, true, true])]; + tensor var_112_cast_fp16 = slice_by_index(begin = var_112_begin_0, end = var_112_end_0, end_mask = var_112_end_mask_0, x = qkv_1_cast_fp16)[name = string("op_112_cast_fp16")]; + tensor var_113 = const()[name = string("op_113"), val = tensor([1, 1024, 32, 128])]; + tensor v_1_cast_fp16 = reshape(shape = var_113, x = var_112_cast_fp16)[name = string("v_1_cast_fp16")]; + tensor layers_0_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_0_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42030400)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63001984)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_0_linear_attn_in_proj_z_weight_to_fp16, x = x_3_cast_fp16)[name = string("linear_1_cast_fp16")]; + tensor var_117 = const()[name = string("op_117"), val = tensor([1, 1024, 32, 128])]; + tensor z_1_cast_fp16 = reshape(shape = var_117, x = linear_1_cast_fp16)[name = string("z_1_cast_fp16")]; + tensor layers_0_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_0_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63010240)))]; + tensor linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63174144)))]; + tensor linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = layers_0_linear_attn_in_proj_b_weight_to_fp16, x = x_3_cast_fp16)[name = string("linear_2_cast_fp16")]; + tensor beta_1_cast_fp16 = sigmoid(x = linear_2_cast_fp16)[name = string("beta_1_cast_fp16")]; + tensor var_126_weight_0_to_fp16 = const()[name = string("op_126_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63174272)))]; + tensor var_126_bias_0_to_fp16 = const()[name = string("op_126_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63338176)))]; + tensor var_126_cast_fp16 = linear(bias = var_126_bias_0_to_fp16, weight = var_126_weight_0_to_fp16, x = x_3_cast_fp16)[name = string("op_126_cast_fp16")]; + tensor var_127_cast_fp16 = softplus(x = var_126_cast_fp16)[name = string("op_127_cast_fp16")]; + tensor var_123_to_fp16 = const()[name = string("op_123_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63338304)))]; + tensor g_1_cast_fp16 = mul(x = var_123_to_fp16, y = var_127_cast_fp16)[name = string("g_1_cast_fp16")]; + tensor var_129_cast_fp16 = mul(x = q_1_cast_fp16, y = q_1_cast_fp16)[name = string("op_129_cast_fp16")]; + tensor var_131_axes_0 = const()[name = string("op_131_axes_0"), val = tensor([-1])]; + bool var_131_keep_dims_0 = const()[name = string("op_131_keep_dims_0"), val = bool(true)]; + tensor var_131_cast_fp16 = reduce_sum(axes = var_131_axes_0, keep_dims = var_131_keep_dims_0, x = var_129_cast_fp16)[name = string("op_131_cast_fp16")]; + fp16 var_132_to_fp16 = const()[name = string("op_132_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_133_cast_fp16 = add(x = var_131_cast_fp16, y = var_132_to_fp16)[name = string("op_133_cast_fp16")]; + fp32 var_134_epsilon_0 = const()[name = string("op_134_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_134_cast_fp16 = rsqrt(epsilon = var_134_epsilon_0, x = var_133_cast_fp16)[name = string("op_134_cast_fp16")]; + tensor q_3_cast_fp16 = mul(x = q_1_cast_fp16, y = var_134_cast_fp16)[name = string("q_3_cast_fp16")]; + tensor var_136_cast_fp16 = mul(x = k_1_cast_fp16, y = k_1_cast_fp16)[name = string("op_136_cast_fp16")]; + tensor var_138_axes_0 = const()[name = string("op_138_axes_0"), val = tensor([-1])]; + bool var_138_keep_dims_0 = const()[name = string("op_138_keep_dims_0"), val = bool(true)]; + tensor var_138_cast_fp16 = reduce_sum(axes = var_138_axes_0, keep_dims = var_138_keep_dims_0, x = var_136_cast_fp16)[name = string("op_138_cast_fp16")]; + fp16 var_139_to_fp16 = const()[name = string("op_139_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_140_cast_fp16 = add(x = var_138_cast_fp16, y = var_139_to_fp16)[name = string("op_140_cast_fp16")]; + fp32 var_141_epsilon_0 = const()[name = string("op_141_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_141_cast_fp16 = rsqrt(epsilon = var_141_epsilon_0, x = var_140_cast_fp16)[name = string("op_141_cast_fp16")]; + tensor k_3_cast_fp16 = mul(x = k_1_cast_fp16, y = var_141_cast_fp16)[name = string("k_3_cast_fp16")]; + tensor var_146_axes_0 = const()[name = string("op_146_axes_0"), val = tensor([3])]; + tensor var_146_cast_fp16 = expand_dims(axes = var_146_axes_0, x = q_3_cast_fp16)[name = string("op_146_cast_fp16")]; + tensor var_148_reps_0 = const()[name = string("op_148_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_148_cast_fp16 = tile(reps = var_148_reps_0, x = var_146_cast_fp16)[name = string("op_148_cast_fp16")]; + tensor var_149 = const()[name = string("op_149"), val = tensor([1, 1024, 32, 128])]; + tensor q_5_cast_fp16 = reshape(shape = var_149, x = var_148_cast_fp16)[name = string("q_5_cast_fp16")]; + tensor var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor([3])]; + tensor var_154_cast_fp16 = expand_dims(axes = var_154_axes_0, x = k_3_cast_fp16)[name = string("op_154_cast_fp16")]; + tensor var_156_reps_0 = const()[name = string("op_156_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_156_cast_fp16 = tile(reps = var_156_reps_0, x = var_154_cast_fp16)[name = string("op_156_cast_fp16")]; + tensor var_157 = const()[name = string("op_157"), val = tensor([1, 1024, 32, 128])]; + tensor k_5_cast_fp16 = reshape(shape = var_157, x = var_156_cast_fp16)[name = string("k_5_cast_fp16")]; + fp16 var_159_to_fp16 = const()[name = string("op_159_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_7_cast_fp16 = mul(x = q_5_cast_fp16, y = var_159_to_fp16)[name = string("q_7_cast_fp16")]; + tensor var_161_begin_0 = const()[name = string("op_161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_161_end_0 = const()[name = string("op_161_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_161_end_mask_0 = const()[name = string("op_161_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_161_squeeze_mask_0 = const()[name = string("op_161_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_161_cast_fp16 = slice_by_index(begin = var_161_begin_0, end = var_161_end_0, end_mask = var_161_end_mask_0, squeeze_mask = var_161_squeeze_mask_0, x = q_7_cast_fp16)[name = string("op_161_cast_fp16")]; + tensor var_162 = const()[name = string("op_162"), val = tensor([1, 0, 2])]; + tensor var_164 = const()[name = string("op_164"), val = tensor([32, 16, 64, 128])]; + tensor var_163_cast_fp16 = transpose(perm = var_162, x = var_161_cast_fp16)[name = string("transpose_53")]; + tensor q_9_cast_fp16 = reshape(shape = var_164, x = var_163_cast_fp16)[name = string("q_9_cast_fp16")]; + tensor var_166_begin_0 = const()[name = string("op_166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_166_end_0 = const()[name = string("op_166_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_166_end_mask_0 = const()[name = string("op_166_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_166_squeeze_mask_0 = const()[name = string("op_166_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_166_cast_fp16 = slice_by_index(begin = var_166_begin_0, end = var_166_end_0, end_mask = var_166_end_mask_0, squeeze_mask = var_166_squeeze_mask_0, x = k_5_cast_fp16)[name = string("op_166_cast_fp16")]; + tensor var_167 = const()[name = string("op_167"), val = tensor([1, 0, 2])]; + tensor var_169 = const()[name = string("op_169"), val = tensor([32, 16, 64, 128])]; + tensor var_168_cast_fp16 = transpose(perm = var_167, x = var_166_cast_fp16)[name = string("transpose_52")]; + tensor k_7_cast_fp16 = reshape(shape = var_169, x = var_168_cast_fp16)[name = string("k_7_cast_fp16")]; + tensor var_171_begin_0 = const()[name = string("op_171_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_171_end_0 = const()[name = string("op_171_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_171_end_mask_0 = const()[name = string("op_171_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_171_squeeze_mask_0 = const()[name = string("op_171_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_171_cast_fp16 = slice_by_index(begin = var_171_begin_0, end = var_171_end_0, end_mask = var_171_end_mask_0, squeeze_mask = var_171_squeeze_mask_0, x = v_1_cast_fp16)[name = string("op_171_cast_fp16")]; + tensor var_172 = const()[name = string("op_172"), val = tensor([1, 0, 2])]; + tensor var_174 = const()[name = string("op_174"), val = tensor([32, 16, 64, 128])]; + tensor var_173_cast_fp16 = transpose(perm = var_172, x = var_171_cast_fp16)[name = string("transpose_51")]; + tensor v_3_cast_fp16 = reshape(shape = var_174, x = var_173_cast_fp16)[name = string("v_3_cast_fp16")]; + tensor var_176_begin_0 = const()[name = string("op_176_begin_0"), val = tensor([0, 0, 0])]; + tensor var_176_end_0 = const()[name = string("op_176_end_0"), val = tensor([1, 1024, 32])]; + tensor var_176_end_mask_0 = const()[name = string("op_176_end_mask_0"), val = tensor([false, true, true])]; + tensor var_176_squeeze_mask_0 = const()[name = string("op_176_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_176_cast_fp16 = slice_by_index(begin = var_176_begin_0, end = var_176_end_0, end_mask = var_176_end_mask_0, squeeze_mask = var_176_squeeze_mask_0, x = beta_1_cast_fp16)[name = string("op_176_cast_fp16")]; + tensor var_177_perm_0 = const()[name = string("op_177_perm_0"), val = tensor([1, 0])]; + tensor var_178 = const()[name = string("op_178"), val = tensor([32, 16, 64, 1])]; + tensor var_177_cast_fp16 = transpose(perm = var_177_perm_0, x = var_176_cast_fp16)[name = string("transpose_50")]; + tensor beta_3_cast_fp16 = reshape(shape = var_178, x = var_177_cast_fp16)[name = string("beta_3_cast_fp16")]; + tensor var_180_begin_0 = const()[name = string("op_180_begin_0"), val = tensor([0, 0, 0])]; + tensor var_180_end_0 = const()[name = string("op_180_end_0"), val = tensor([1, 1024, 32])]; + tensor var_180_end_mask_0 = const()[name = string("op_180_end_mask_0"), val = tensor([false, true, true])]; + tensor var_180_squeeze_mask_0 = const()[name = string("op_180_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_180_cast_fp16 = slice_by_index(begin = var_180_begin_0, end = var_180_end_0, end_mask = var_180_end_mask_0, squeeze_mask = var_180_squeeze_mask_0, x = g_1_cast_fp16)[name = string("op_180_cast_fp16")]; + tensor var_181_perm_0 = const()[name = string("op_181_perm_0"), val = tensor([1, 0])]; + tensor var_182 = const()[name = string("op_182"), val = tensor([32, 16, 64])]; + tensor var_181_cast_fp16 = transpose(perm = var_181_perm_0, x = var_180_cast_fp16)[name = string("transpose_49")]; + tensor g_3_cast_fp16 = reshape(shape = var_182, x = var_181_cast_fp16)[name = string("g_3_cast_fp16")]; + tensor g_col_1_axes_0 = const()[name = string("g_col_1_axes_0"), val = tensor([-1])]; + tensor g_col_1_cast_fp16 = expand_dims(axes = g_col_1_axes_0, x = g_3_cast_fp16)[name = string("g_col_1_cast_fp16")]; + bool var_185_transpose_x_0 = const()[name = string("op_185_transpose_x_0"), val = bool(false)]; + bool var_185_transpose_y_0 = const()[name = string("op_185_transpose_y_0"), val = bool(false)]; + tensor layers_0_linear_attn_lower_incl_to_fp16 = const()[name = string("layers_0_linear_attn_lower_incl_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63338432)))]; + tensor var_185_cast_fp16 = matmul(transpose_x = var_185_transpose_x_0, transpose_y = var_185_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_1_cast_fp16)[name = string("op_185_cast_fp16")]; + tensor cum_1_axes_0 = const()[name = string("cum_1_axes_0"), val = tensor([-1])]; + tensor cum_1_cast_fp16 = squeeze(axes = cum_1_axes_0, x = var_185_cast_fp16)[name = string("cum_1_cast_fp16")]; + bool var_187_transpose_x_0 = const()[name = string("op_187_transpose_x_0"), val = bool(false)]; + bool var_187_transpose_y_0 = const()[name = string("op_187_transpose_y_0"), val = bool(false)]; + tensor layers_0_linear_attn_suffix_sum_t_to_fp16 = const()[name = string("layers_0_linear_attn_suffix_sum_t_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63346688)))]; + tensor var_187_cast_fp16 = matmul(transpose_x = var_187_transpose_x_0, transpose_y = var_187_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_1_cast_fp16)[name = string("op_187_cast_fp16")]; + tensor to_end_1_axes_0 = const()[name = string("to_end_1_axes_0"), val = tensor([-1])]; + tensor to_end_1_cast_fp16 = squeeze(axes = to_end_1_axes_0, x = var_187_cast_fp16)[name = string("to_end_1_cast_fp16")]; + bool var_189_transpose_x_0 = const()[name = string("op_189_transpose_x_0"), val = bool(false)]; + bool var_189_transpose_y_0 = const()[name = string("op_189_transpose_y_0"), val = bool(false)]; + tensor layers_0_linear_attn_pair_sum_t_to_fp16 = const()[name = string("layers_0_linear_attn_pair_sum_t_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63354944)))]; + tensor var_189_cast_fp16 = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_1_cast_fp16)[name = string("op_189_cast_fp16")]; + tensor var_190 = const()[name = string("op_190"), val = tensor([32, 16, 64, 64])]; + tensor pair_1_cast_fp16 = reshape(shape = var_190, x = var_189_cast_fp16)[name = string("pair_1_cast_fp16")]; + tensor var_192_cast_fp16 = exp(x = pair_1_cast_fp16)[name = string("op_192_cast_fp16")]; + tensor pair_decay_1_cast_fp16 = mul(x = var_192_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_1_cast_fp16")]; + tensor k_beta_1_cast_fp16 = mul(x = k_7_cast_fp16, y = beta_3_cast_fp16)[name = string("k_beta_1_cast_fp16")]; + tensor v_beta_1_cast_fp16 = mul(x = v_3_cast_fp16, y = beta_3_cast_fp16)[name = string("v_beta_1_cast_fp16")]; + bool var_197_transpose_x_1 = const()[name = string("op_197_transpose_x_1"), val = bool(false)]; + bool var_197_transpose_y_1 = const()[name = string("op_197_transpose_y_1"), val = bool(true)]; + tensor var_197_cast_fp16 = matmul(transpose_x = var_197_transpose_x_1, transpose_y = var_197_transpose_y_1, x = k_beta_1_cast_fp16, y = k_7_cast_fp16)[name = string("op_197_cast_fp16")]; + tensor kkt_1_cast_fp16 = mul(x = var_197_cast_fp16, y = pair_decay_1_cast_fp16)[name = string("kkt_1_cast_fp16")]; + bool var_200_transpose_x_1 = const()[name = string("op_200_transpose_x_1"), val = bool(false)]; + bool var_200_transpose_y_1 = const()[name = string("op_200_transpose_y_1"), val = bool(true)]; + tensor var_200_cast_fp16 = matmul(transpose_x = var_200_transpose_x_1, transpose_y = var_200_transpose_y_1, x = q_9_cast_fp16, y = k_7_cast_fp16)[name = string("op_200_cast_fp16")]; + tensor attn_intra_1_cast_fp16 = mul(x = var_200_cast_fp16, y = pair_decay_1_cast_fp16)[name = string("attn_intra_1_cast_fp16")]; + tensor layers_0_linear_attn_strict_lower_to_fp16 = const()[name = string("layers_0_linear_attn_strict_lower_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63879296)))]; + tensor var_202_cast_fp16 = mul(x = kkt_1_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_202_cast_fp16")]; + tensor layers_0_linear_attn_eye_to_fp16 = const()[name = string("layers_0_linear_attn_eye_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63887552)))]; + tensor t_1_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_202_cast_fp16)[name = string("t_1_cast_fp16")]; + tensor var_206 = const()[name = string("op_206"), val = tensor([-1, 64, 64])]; + tensor t_3_cast_fp16 = reshape(shape = var_206, x = t_1_cast_fp16)[name = string("t_3_cast_fp16")]; + tensor var_211 = const()[name = string("op_211"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_212_cast_fp16 = reshape(shape = var_211, x = t_3_cast_fp16)[name = string("op_212_cast_fp16")]; + tensor var_216_to_fp16 = const()[name = string("op_216_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63895808)))]; + tensor var_217_cast_fp16 = mul(x = var_212_cast_fp16, y = var_216_to_fp16)[name = string("op_217_cast_fp16")]; + tensor blocks_1_axes_0 = const()[name = string("blocks_1_axes_0"), val = tensor([-2])]; + bool blocks_1_keep_dims_0 = const()[name = string("blocks_1_keep_dims_0"), val = bool(false)]; + tensor blocks_1_cast_fp16 = reduce_sum(axes = blocks_1_axes_0, keep_dims = blocks_1_keep_dims_0, x = var_217_cast_fp16)[name = string("blocks_1_cast_fp16")]; + tensor var_220_begin_0 = const()[name = string("op_220_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_220_end_0 = const()[name = string("op_220_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_220_end_mask_0 = const()[name = string("op_220_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_220_cast_fp16 = slice_by_index(begin = var_220_begin_0, end = var_220_end_0, end_mask = var_220_end_mask_0, x = blocks_1_cast_fp16)[name = string("op_220_cast_fp16")]; + tensor x_5_begin_0 = const()[name = string("x_5_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_5_end_0 = const()[name = string("x_5_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_5_end_mask_0 = const()[name = string("x_5_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_5_cast_fp16 = slice_by_index(begin = x_5_begin_0, end = x_5_end_0, end_mask = x_5_end_mask_0, x = var_220_cast_fp16)[name = string("x_5_cast_fp16")]; + bool var_230_transpose_x_0 = const()[name = string("op_230_transpose_x_0"), val = bool(false)]; + bool var_230_transpose_y_0 = const()[name = string("op_230_transpose_y_0"), val = bool(false)]; + tensor var_224_to_fp16 = const()[name = string("op_224_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63897920)))]; + tensor var_230_cast_fp16 = matmul(transpose_x = var_230_transpose_x_0, transpose_y = var_230_transpose_y_0, x = x_5_cast_fp16, y = var_224_to_fp16)[name = string("op_230_cast_fp16")]; + bool var_231_transpose_x_0 = const()[name = string("op_231_transpose_x_0"), val = bool(false)]; + bool var_231_transpose_y_0 = const()[name = string("op_231_transpose_y_0"), val = bool(false)]; + tensor var_231_cast_fp16 = matmul(transpose_x = var_231_transpose_x_0, transpose_y = var_231_transpose_y_0, x = var_224_to_fp16, y = var_230_cast_fp16)[name = string("op_231_cast_fp16")]; + fp16 const_6_promoted_to_fp16 = const()[name = string("const_6_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_1_cast_fp16 = mul(x = var_231_cast_fp16, y = const_6_promoted_to_fp16)[name = string("lower_1_cast_fp16")]; + bool bottom_1_interleave_0 = const()[name = string("bottom_1_interleave_0"), val = bool(false)]; + tensor bottom_1_cast_fp16 = concat(axis = var_58, interleave = bottom_1_interleave_0, values = (lower_1_cast_fp16, var_224_to_fp16))[name = string("bottom_1_cast_fp16")]; + bool inv_3_interleave_0 = const()[name = string("inv_3_interleave_0"), val = bool(false)]; + tensor top_1_to_fp16 = const()[name = string("top_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63930752)))]; + tensor inv_3_cast_fp16 = concat(axis = var_37, interleave = inv_3_interleave_0, values = (top_1_to_fp16, bottom_1_cast_fp16))[name = string("inv_3_cast_fp16")]; + tensor var_240 = const()[name = string("op_240"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_241_cast_fp16 = reshape(shape = var_240, x = t_3_cast_fp16)[name = string("op_241_cast_fp16")]; + tensor var_245_to_fp16 = const()[name = string("op_245_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63996352)))]; + tensor var_246_cast_fp16 = mul(x = var_241_cast_fp16, y = var_245_to_fp16)[name = string("op_246_cast_fp16")]; + tensor blocks_3_axes_0 = const()[name = string("blocks_3_axes_0"), val = tensor([-2])]; + bool blocks_3_keep_dims_0 = const()[name = string("blocks_3_keep_dims_0"), val = bool(false)]; + tensor blocks_3_cast_fp16 = reduce_sum(axes = blocks_3_axes_0, keep_dims = blocks_3_keep_dims_0, x = var_246_cast_fp16)[name = string("blocks_3_cast_fp16")]; + tensor var_249_begin_0 = const()[name = string("op_249_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_249_end_0 = const()[name = string("op_249_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_249_end_mask_0 = const()[name = string("op_249_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_249_cast_fp16 = slice_by_index(begin = var_249_begin_0, end = var_249_end_0, end_mask = var_249_end_mask_0, x = blocks_3_cast_fp16)[name = string("op_249_cast_fp16")]; + tensor x_7_begin_0 = const()[name = string("x_7_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_7_end_0 = const()[name = string("x_7_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_7_end_mask_0 = const()[name = string("x_7_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_7_cast_fp16 = slice_by_index(begin = x_7_begin_0, end = x_7_end_0, end_mask = x_7_end_mask_0, x = var_249_cast_fp16)[name = string("x_7_cast_fp16")]; + tensor var_251 = const()[name = string("op_251"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_5_cast_fp16 = reshape(shape = var_251, x = inv_3_cast_fp16)[name = string("pair_5_cast_fp16")]; + tensor var_253_begin_0 = const()[name = string("op_253_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_253_end_0 = const()[name = string("op_253_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_253_end_mask_0 = const()[name = string("op_253_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_253_squeeze_mask_0 = const()[name = string("op_253_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_253_cast_fp16 = slice_by_index(begin = var_253_begin_0, end = var_253_end_0, end_mask = var_253_end_mask_0, squeeze_mask = var_253_squeeze_mask_0, x = pair_5_cast_fp16)[name = string("op_253_cast_fp16")]; + tensor var_256_begin_0 = const()[name = string("op_256_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_256_end_0 = const()[name = string("op_256_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_256_end_mask_0 = const()[name = string("op_256_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_256_squeeze_mask_0 = const()[name = string("op_256_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_256_cast_fp16 = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, squeeze_mask = var_256_squeeze_mask_0, x = pair_5_cast_fp16)[name = string("op_256_cast_fp16")]; + bool var_259_transpose_x_0 = const()[name = string("op_259_transpose_x_0"), val = bool(false)]; + bool var_259_transpose_y_0 = const()[name = string("op_259_transpose_y_0"), val = bool(false)]; + tensor var_259_cast_fp16 = matmul(transpose_x = var_259_transpose_x_0, transpose_y = var_259_transpose_y_0, x = x_7_cast_fp16, y = var_253_cast_fp16)[name = string("op_259_cast_fp16")]; + bool var_260_transpose_x_0 = const()[name = string("op_260_transpose_x_0"), val = bool(false)]; + bool var_260_transpose_y_0 = const()[name = string("op_260_transpose_y_0"), val = bool(false)]; + tensor var_260_cast_fp16 = matmul(transpose_x = var_260_transpose_x_0, transpose_y = var_260_transpose_y_0, x = var_256_cast_fp16, y = var_259_cast_fp16)[name = string("op_260_cast_fp16")]; + fp16 const_7_promoted_to_fp16 = const()[name = string("const_7_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_3_cast_fp16 = mul(x = var_260_cast_fp16, y = const_7_promoted_to_fp16)[name = string("lower_3_cast_fp16")]; + bool top_3_interleave_0 = const()[name = string("top_3_interleave_0"), val = bool(false)]; + tensor var_262_to_fp16 = const()[name = string("op_262_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63996928)))]; + tensor top_3_cast_fp16 = concat(axis = var_58, interleave = top_3_interleave_0, values = (var_253_cast_fp16, var_262_to_fp16))[name = string("top_3_cast_fp16")]; + bool bottom_3_interleave_0 = const()[name = string("bottom_3_interleave_0"), val = bool(false)]; + tensor bottom_3_cast_fp16 = concat(axis = var_58, interleave = bottom_3_interleave_0, values = (lower_3_cast_fp16, var_256_cast_fp16))[name = string("bottom_3_cast_fp16")]; + bool inv_5_interleave_0 = const()[name = string("inv_5_interleave_0"), val = bool(false)]; + tensor inv_5_cast_fp16 = concat(axis = var_37, interleave = inv_5_interleave_0, values = (top_3_cast_fp16, bottom_3_cast_fp16))[name = string("inv_5_cast_fp16")]; + tensor var_269 = const()[name = string("op_269"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_270_cast_fp16 = reshape(shape = var_269, x = t_3_cast_fp16)[name = string("op_270_cast_fp16")]; + tensor var_274_to_fp16 = const()[name = string("op_274_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64062528)))]; + tensor var_275_cast_fp16 = mul(x = var_270_cast_fp16, y = var_274_to_fp16)[name = string("op_275_cast_fp16")]; + tensor blocks_5_axes_0 = const()[name = string("blocks_5_axes_0"), val = tensor([-2])]; + bool blocks_5_keep_dims_0 = const()[name = string("blocks_5_keep_dims_0"), val = bool(false)]; + tensor blocks_5_cast_fp16 = reduce_sum(axes = blocks_5_axes_0, keep_dims = blocks_5_keep_dims_0, x = var_275_cast_fp16)[name = string("blocks_5_cast_fp16")]; + tensor var_278_begin_0 = const()[name = string("op_278_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_278_end_0 = const()[name = string("op_278_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_278_end_mask_0 = const()[name = string("op_278_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_278_cast_fp16 = slice_by_index(begin = var_278_begin_0, end = var_278_end_0, end_mask = var_278_end_mask_0, x = blocks_5_cast_fp16)[name = string("op_278_cast_fp16")]; + tensor x_9_begin_0 = const()[name = string("x_9_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_9_end_0 = const()[name = string("x_9_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_9_end_mask_0 = const()[name = string("x_9_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_9_cast_fp16 = slice_by_index(begin = x_9_begin_0, end = x_9_end_0, end_mask = x_9_end_mask_0, x = var_278_cast_fp16)[name = string("x_9_cast_fp16")]; + tensor var_280 = const()[name = string("op_280"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_7_cast_fp16 = reshape(shape = var_280, x = inv_5_cast_fp16)[name = string("pair_7_cast_fp16")]; + tensor var_282_begin_0 = const()[name = string("op_282_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_282_end_0 = const()[name = string("op_282_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_282_end_mask_0 = const()[name = string("op_282_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_282_squeeze_mask_0 = const()[name = string("op_282_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_282_cast_fp16 = slice_by_index(begin = var_282_begin_0, end = var_282_end_0, end_mask = var_282_end_mask_0, squeeze_mask = var_282_squeeze_mask_0, x = pair_7_cast_fp16)[name = string("op_282_cast_fp16")]; + tensor var_285_begin_0 = const()[name = string("op_285_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_285_end_0 = const()[name = string("op_285_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_285_end_mask_0 = const()[name = string("op_285_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_285_squeeze_mask_0 = const()[name = string("op_285_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_285_cast_fp16 = slice_by_index(begin = var_285_begin_0, end = var_285_end_0, end_mask = var_285_end_mask_0, squeeze_mask = var_285_squeeze_mask_0, x = pair_7_cast_fp16)[name = string("op_285_cast_fp16")]; + bool var_288_transpose_x_0 = const()[name = string("op_288_transpose_x_0"), val = bool(false)]; + bool var_288_transpose_y_0 = const()[name = string("op_288_transpose_y_0"), val = bool(false)]; + tensor var_288_cast_fp16 = matmul(transpose_x = var_288_transpose_x_0, transpose_y = var_288_transpose_y_0, x = x_9_cast_fp16, y = var_282_cast_fp16)[name = string("op_288_cast_fp16")]; + bool var_289_transpose_x_0 = const()[name = string("op_289_transpose_x_0"), val = bool(false)]; + bool var_289_transpose_y_0 = const()[name = string("op_289_transpose_y_0"), val = bool(false)]; + tensor var_289_cast_fp16 = matmul(transpose_x = var_289_transpose_x_0, transpose_y = var_289_transpose_y_0, x = var_285_cast_fp16, y = var_288_cast_fp16)[name = string("op_289_cast_fp16")]; + fp16 const_8_promoted_to_fp16 = const()[name = string("const_8_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_5_cast_fp16 = mul(x = var_289_cast_fp16, y = const_8_promoted_to_fp16)[name = string("lower_5_cast_fp16")]; + bool top_5_interleave_0 = const()[name = string("top_5_interleave_0"), val = bool(false)]; + tensor var_291_to_fp16 = const()[name = string("op_291_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64062720)))]; + tensor top_5_cast_fp16 = concat(axis = var_58, interleave = top_5_interleave_0, values = (var_282_cast_fp16, var_291_to_fp16))[name = string("top_5_cast_fp16")]; + bool bottom_5_interleave_0 = const()[name = string("bottom_5_interleave_0"), val = bool(false)]; + tensor bottom_5_cast_fp16 = concat(axis = var_58, interleave = bottom_5_interleave_0, values = (lower_5_cast_fp16, var_285_cast_fp16))[name = string("bottom_5_cast_fp16")]; + bool inv_7_interleave_0 = const()[name = string("inv_7_interleave_0"), val = bool(false)]; + tensor inv_7_cast_fp16 = concat(axis = var_37, interleave = inv_7_interleave_0, values = (top_5_cast_fp16, bottom_5_cast_fp16))[name = string("inv_7_cast_fp16")]; + tensor var_298 = const()[name = string("op_298"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_299_cast_fp16 = reshape(shape = var_298, x = t_3_cast_fp16)[name = string("op_299_cast_fp16")]; + tensor var_303_to_fp16 = const()[name = string("op_303_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64193856)))]; + tensor var_304_cast_fp16 = mul(x = var_299_cast_fp16, y = var_303_to_fp16)[name = string("op_304_cast_fp16")]; + tensor blocks_7_axes_0 = const()[name = string("blocks_7_axes_0"), val = tensor([-2])]; + bool blocks_7_keep_dims_0 = const()[name = string("blocks_7_keep_dims_0"), val = bool(false)]; + tensor blocks_7_cast_fp16 = reduce_sum(axes = blocks_7_axes_0, keep_dims = blocks_7_keep_dims_0, x = var_304_cast_fp16)[name = string("blocks_7_cast_fp16")]; + tensor var_307_begin_0 = const()[name = string("op_307_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_307_end_0 = const()[name = string("op_307_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_307_end_mask_0 = const()[name = string("op_307_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_307_cast_fp16 = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = blocks_7_cast_fp16)[name = string("op_307_cast_fp16")]; + tensor x_11_begin_0 = const()[name = string("x_11_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_11_end_0 = const()[name = string("x_11_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_11_end_mask_0 = const()[name = string("x_11_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_11_cast_fp16 = slice_by_index(begin = x_11_begin_0, end = x_11_end_0, end_mask = x_11_end_mask_0, x = var_307_cast_fp16)[name = string("x_11_cast_fp16")]; + tensor var_309 = const()[name = string("op_309"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_9_cast_fp16 = reshape(shape = var_309, x = inv_7_cast_fp16)[name = string("pair_9_cast_fp16")]; + tensor var_311_begin_0 = const()[name = string("op_311_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_311_end_0 = const()[name = string("op_311_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_311_end_mask_0 = const()[name = string("op_311_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_311_squeeze_mask_0 = const()[name = string("op_311_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_311_cast_fp16 = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, squeeze_mask = var_311_squeeze_mask_0, x = pair_9_cast_fp16)[name = string("op_311_cast_fp16")]; + tensor var_314_begin_0 = const()[name = string("op_314_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_314_end_0 = const()[name = string("op_314_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_314_end_mask_0 = const()[name = string("op_314_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_314_squeeze_mask_0 = const()[name = string("op_314_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_314_cast_fp16 = slice_by_index(begin = var_314_begin_0, end = var_314_end_0, end_mask = var_314_end_mask_0, squeeze_mask = var_314_squeeze_mask_0, x = pair_9_cast_fp16)[name = string("op_314_cast_fp16")]; + bool var_317_transpose_x_0 = const()[name = string("op_317_transpose_x_0"), val = bool(false)]; + bool var_317_transpose_y_0 = const()[name = string("op_317_transpose_y_0"), val = bool(false)]; + tensor var_317_cast_fp16 = matmul(transpose_x = var_317_transpose_x_0, transpose_y = var_317_transpose_y_0, x = x_11_cast_fp16, y = var_311_cast_fp16)[name = string("op_317_cast_fp16")]; + bool var_318_transpose_x_0 = const()[name = string("op_318_transpose_x_0"), val = bool(false)]; + bool var_318_transpose_y_0 = const()[name = string("op_318_transpose_y_0"), val = bool(false)]; + tensor var_318_cast_fp16 = matmul(transpose_x = var_318_transpose_x_0, transpose_y = var_318_transpose_y_0, x = var_314_cast_fp16, y = var_317_cast_fp16)[name = string("op_318_cast_fp16")]; + fp16 const_9_promoted_to_fp16 = const()[name = string("const_9_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_7_cast_fp16 = mul(x = var_318_cast_fp16, y = const_9_promoted_to_fp16)[name = string("lower_7_cast_fp16")]; + bool top_7_interleave_0 = const()[name = string("top_7_interleave_0"), val = bool(false)]; + tensor var_320_to_fp16 = const()[name = string("op_320_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64193984)))]; + tensor top_7_cast_fp16 = concat(axis = var_58, interleave = top_7_interleave_0, values = (var_311_cast_fp16, var_320_to_fp16))[name = string("top_7_cast_fp16")]; + bool bottom_7_interleave_0 = const()[name = string("bottom_7_interleave_0"), val = bool(false)]; + tensor bottom_7_cast_fp16 = concat(axis = var_58, interleave = bottom_7_interleave_0, values = (lower_7_cast_fp16, var_314_cast_fp16))[name = string("bottom_7_cast_fp16")]; + bool inv_9_interleave_0 = const()[name = string("inv_9_interleave_0"), val = bool(false)]; + tensor inv_9_cast_fp16 = concat(axis = var_37, interleave = inv_9_interleave_0, values = (top_7_cast_fp16, bottom_7_cast_fp16))[name = string("inv_9_cast_fp16")]; + tensor var_327 = const()[name = string("op_327"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_328_cast_fp16 = reshape(shape = var_327, x = t_3_cast_fp16)[name = string("op_328_cast_fp16")]; + tensor var_332_to_fp16 = const()[name = string("op_332_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_333_cast_fp16 = mul(x = var_328_cast_fp16, y = var_332_to_fp16)[name = string("op_333_cast_fp16")]; + tensor blocks_9_axes_0 = const()[name = string("blocks_9_axes_0"), val = tensor([-2])]; + bool blocks_9_keep_dims_0 = const()[name = string("blocks_9_keep_dims_0"), val = bool(false)]; + tensor blocks_9_cast_fp16 = reduce_sum(axes = blocks_9_axes_0, keep_dims = blocks_9_keep_dims_0, x = var_333_cast_fp16)[name = string("blocks_9_cast_fp16")]; + tensor var_336_begin_0 = const()[name = string("op_336_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_336_end_0 = const()[name = string("op_336_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_336_end_mask_0 = const()[name = string("op_336_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_336_cast_fp16 = slice_by_index(begin = var_336_begin_0, end = var_336_end_0, end_mask = var_336_end_mask_0, x = blocks_9_cast_fp16)[name = string("op_336_cast_fp16")]; + tensor x_13_begin_0 = const()[name = string("x_13_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_13_end_0 = const()[name = string("x_13_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_13_end_mask_0 = const()[name = string("x_13_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_13_cast_fp16 = slice_by_index(begin = x_13_begin_0, end = x_13_end_0, end_mask = x_13_end_mask_0, x = var_336_cast_fp16)[name = string("x_13_cast_fp16")]; + tensor var_338 = const()[name = string("op_338"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_11_cast_fp16 = reshape(shape = var_338, x = inv_9_cast_fp16)[name = string("pair_11_cast_fp16")]; + tensor var_340_begin_0 = const()[name = string("op_340_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_340_end_0 = const()[name = string("op_340_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_340_end_mask_0 = const()[name = string("op_340_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_340_squeeze_mask_0 = const()[name = string("op_340_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_340_cast_fp16 = slice_by_index(begin = var_340_begin_0, end = var_340_end_0, end_mask = var_340_end_mask_0, squeeze_mask = var_340_squeeze_mask_0, x = pair_11_cast_fp16)[name = string("op_340_cast_fp16")]; + tensor var_343_begin_0 = const()[name = string("op_343_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_343_end_0 = const()[name = string("op_343_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_343_end_mask_0 = const()[name = string("op_343_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_343_squeeze_mask_0 = const()[name = string("op_343_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_343_cast_fp16 = slice_by_index(begin = var_343_begin_0, end = var_343_end_0, end_mask = var_343_end_mask_0, squeeze_mask = var_343_squeeze_mask_0, x = pair_11_cast_fp16)[name = string("op_343_cast_fp16")]; + bool var_346_transpose_x_0 = const()[name = string("op_346_transpose_x_0"), val = bool(false)]; + bool var_346_transpose_y_0 = const()[name = string("op_346_transpose_y_0"), val = bool(false)]; + tensor var_346_cast_fp16 = matmul(transpose_x = var_346_transpose_x_0, transpose_y = var_346_transpose_y_0, x = x_13_cast_fp16, y = var_340_cast_fp16)[name = string("op_346_cast_fp16")]; + bool var_347_transpose_x_0 = const()[name = string("op_347_transpose_x_0"), val = bool(false)]; + bool var_347_transpose_y_0 = const()[name = string("op_347_transpose_y_0"), val = bool(false)]; + tensor var_347_cast_fp16 = matmul(transpose_x = var_347_transpose_x_0, transpose_y = var_347_transpose_y_0, x = var_343_cast_fp16, y = var_346_cast_fp16)[name = string("op_347_cast_fp16")]; + fp16 const_10_promoted_to_fp16 = const()[name = string("const_10_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_9_cast_fp16 = mul(x = var_347_cast_fp16, y = const_10_promoted_to_fp16)[name = string("lower_9_cast_fp16")]; + bool top_9_interleave_0 = const()[name = string("top_9_interleave_0"), val = bool(false)]; + tensor var_349_to_fp16 = const()[name = string("op_349_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64456192)))]; + tensor top_9_cast_fp16 = concat(axis = var_58, interleave = top_9_interleave_0, values = (var_340_cast_fp16, var_349_to_fp16))[name = string("top_9_cast_fp16")]; + bool bottom_9_interleave_0 = const()[name = string("bottom_9_interleave_0"), val = bool(false)]; + tensor bottom_9_cast_fp16 = concat(axis = var_58, interleave = bottom_9_interleave_0, values = (lower_9_cast_fp16, var_343_cast_fp16))[name = string("bottom_9_cast_fp16")]; + bool inv_11_interleave_0 = const()[name = string("inv_11_interleave_0"), val = bool(false)]; + tensor inv_11_cast_fp16 = concat(axis = var_37, interleave = inv_11_interleave_0, values = (top_9_cast_fp16, bottom_9_cast_fp16))[name = string("inv_11_cast_fp16")]; + tensor var_356 = const()[name = string("op_356"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_357_cast_fp16 = reshape(shape = var_356, x = t_3_cast_fp16)[name = string("op_357_cast_fp16")]; + tensor blocks_11_axes_0 = const()[name = string("blocks_11_axes_0"), val = tensor([-2])]; + bool blocks_11_keep_dims_0 = const()[name = string("blocks_11_keep_dims_0"), val = bool(false)]; + tensor blocks_11_cast_fp16 = reduce_sum(axes = blocks_11_axes_0, keep_dims = blocks_11_keep_dims_0, x = var_357_cast_fp16)[name = string("blocks_11_cast_fp16")]; + tensor var_365_begin_0 = const()[name = string("op_365_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_365_end_0 = const()[name = string("op_365_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_365_end_mask_0 = const()[name = string("op_365_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_365_cast_fp16 = slice_by_index(begin = var_365_begin_0, end = var_365_end_0, end_mask = var_365_end_mask_0, x = blocks_11_cast_fp16)[name = string("op_365_cast_fp16")]; + tensor x_15_begin_0 = const()[name = string("x_15_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_15_end_0 = const()[name = string("x_15_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_15_end_mask_0 = const()[name = string("x_15_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_15_cast_fp16 = slice_by_index(begin = x_15_begin_0, end = x_15_end_0, end_mask = x_15_end_mask_0, x = var_365_cast_fp16)[name = string("x_15_cast_fp16")]; + tensor var_367 = const()[name = string("op_367"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_13_cast_fp16 = reshape(shape = var_367, x = inv_11_cast_fp16)[name = string("pair_13_cast_fp16")]; + tensor var_369_begin_0 = const()[name = string("op_369_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_369_end_0 = const()[name = string("op_369_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_369_end_mask_0 = const()[name = string("op_369_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_369_squeeze_mask_0 = const()[name = string("op_369_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_369_cast_fp16 = slice_by_index(begin = var_369_begin_0, end = var_369_end_0, end_mask = var_369_end_mask_0, squeeze_mask = var_369_squeeze_mask_0, x = pair_13_cast_fp16)[name = string("op_369_cast_fp16")]; + tensor var_372_begin_0 = const()[name = string("op_372_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_372_end_0 = const()[name = string("op_372_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_372_end_mask_0 = const()[name = string("op_372_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_372_squeeze_mask_0 = const()[name = string("op_372_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, squeeze_mask = var_372_squeeze_mask_0, x = pair_13_cast_fp16)[name = string("op_372_cast_fp16")]; + bool var_375_transpose_x_0 = const()[name = string("op_375_transpose_x_0"), val = bool(false)]; + bool var_375_transpose_y_0 = const()[name = string("op_375_transpose_y_0"), val = bool(false)]; + tensor var_375_cast_fp16 = matmul(transpose_x = var_375_transpose_x_0, transpose_y = var_375_transpose_y_0, x = x_15_cast_fp16, y = var_369_cast_fp16)[name = string("op_375_cast_fp16")]; + bool var_376_transpose_x_0 = const()[name = string("op_376_transpose_x_0"), val = bool(false)]; + bool var_376_transpose_y_0 = const()[name = string("op_376_transpose_y_0"), val = bool(false)]; + tensor var_376_cast_fp16 = matmul(transpose_x = var_376_transpose_x_0, transpose_y = var_376_transpose_y_0, x = var_372_cast_fp16, y = var_375_cast_fp16)[name = string("op_376_cast_fp16")]; + fp16 const_11_promoted_to_fp16 = const()[name = string("const_11_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_11_cast_fp16 = mul(x = var_376_cast_fp16, y = const_11_promoted_to_fp16)[name = string("lower_11_cast_fp16")]; + bool top_11_interleave_0 = const()[name = string("top_11_interleave_0"), val = bool(false)]; + tensor var_378_to_fp16 = const()[name = string("op_378_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64980544)))]; + tensor top_11_cast_fp16 = concat(axis = var_58, interleave = top_11_interleave_0, values = (var_369_cast_fp16, var_378_to_fp16))[name = string("top_11_cast_fp16")]; + bool bottom_11_interleave_0 = const()[name = string("bottom_11_interleave_0"), val = bool(false)]; + tensor bottom_11_cast_fp16 = concat(axis = var_58, interleave = bottom_11_interleave_0, values = (lower_11_cast_fp16, var_372_cast_fp16))[name = string("bottom_11_cast_fp16")]; + bool inv_13_interleave_0 = const()[name = string("inv_13_interleave_0"), val = bool(false)]; + tensor inv_13_cast_fp16 = concat(axis = var_37, interleave = inv_13_interleave_0, values = (top_11_cast_fp16, bottom_11_cast_fp16))[name = string("inv_13_cast_fp16")]; + tensor var_385 = const()[name = string("op_385"), val = tensor([32, 16, 64, 64])]; + tensor inv_15_cast_fp16 = reshape(shape = var_385, x = inv_13_cast_fp16)[name = string("inv_15_cast_fp16")]; + bool new_v_1_transpose_x_0 = const()[name = string("new_v_1_transpose_x_0"), val = bool(false)]; + bool new_v_1_transpose_y_0 = const()[name = string("new_v_1_transpose_y_0"), val = bool(false)]; + tensor new_v_1_cast_fp16 = matmul(transpose_x = new_v_1_transpose_x_0, transpose_y = new_v_1_transpose_y_0, x = inv_15_cast_fp16, y = v_beta_1_cast_fp16)[name = string("new_v_1_cast_fp16")]; + tensor var_388_cast_fp16 = exp(x = cum_1_cast_fp16)[name = string("op_388_cast_fp16")]; + tensor var_389_axes_0 = const()[name = string("op_389_axes_0"), val = tensor([3])]; + tensor var_389_cast_fp16 = expand_dims(axes = var_389_axes_0, x = var_388_cast_fp16)[name = string("op_389_cast_fp16")]; + tensor var_390_cast_fp16 = mul(x = k_beta_1_cast_fp16, y = var_389_cast_fp16)[name = string("op_390_cast_fp16")]; + bool k_cumdecay_1_transpose_x_0 = const()[name = string("k_cumdecay_1_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_1_transpose_y_0 = const()[name = string("k_cumdecay_1_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_1_cast_fp16 = matmul(transpose_x = k_cumdecay_1_transpose_x_0, transpose_y = k_cumdecay_1_transpose_y_0, x = inv_15_cast_fp16, y = var_390_cast_fp16)[name = string("k_cumdecay_1_cast_fp16")]; + tensor q_dec_1_cast_fp16 = mul(x = q_9_cast_fp16, y = var_389_cast_fp16)[name = string("q_dec_1_cast_fp16")]; + tensor var_395_cast_fp16 = exp(x = to_end_1_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_396_axes_0 = const()[name = string("op_396_axes_0"), val = tensor([3])]; + tensor var_396_cast_fp16 = expand_dims(axes = var_396_axes_0, x = var_395_cast_fp16)[name = string("op_396_cast_fp16")]; + tensor k_dec_1_cast_fp16 = mul(x = k_7_cast_fp16, y = var_396_cast_fp16)[name = string("k_dec_1_cast_fp16")]; + tensor var_398_begin_0 = const()[name = string("op_398_begin_0"), val = tensor([0, 0, -1])]; + tensor var_398_end_0 = const()[name = string("op_398_end_0"), val = tensor([32, 16, 64])]; + tensor var_398_end_mask_0 = const()[name = string("op_398_end_mask_0"), val = tensor([true, true, true])]; + tensor var_398_cast_fp16 = slice_by_index(begin = var_398_begin_0, end = var_398_end_0, end_mask = var_398_end_mask_0, x = cum_1_cast_fp16)[name = string("op_398_cast_fp16")]; + tensor var_399_cast_fp16 = exp(x = var_398_cast_fp16)[name = string("op_399_cast_fp16")]; + tensor chunk_decay_1_axes_0 = const()[name = string("chunk_decay_1_axes_0"), val = tensor([3])]; + tensor chunk_decay_1_cast_fp16 = expand_dims(axes = chunk_decay_1_axes_0, x = var_399_cast_fp16)[name = string("chunk_decay_1_cast_fp16")]; + tensor var_404_begin_0 = const()[name = string("op_404_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_404_end_0 = const()[name = string("op_404_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_404_end_mask_0 = const()[name = string("op_404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_404_squeeze_mask_0 = const()[name = string("op_404_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_404_cast_fp16 = slice_by_index(begin = var_404_begin_0, end = var_404_end_0, end_mask = var_404_end_mask_0, squeeze_mask = var_404_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_404_cast_fp16")]; + tensor var_406_begin_0 = const()[name = string("op_406_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_406_end_0 = const()[name = string("op_406_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_406_end_mask_0 = const()[name = string("op_406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_406_squeeze_mask_0 = const()[name = string("op_406_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_406_cast_fp16 = slice_by_index(begin = var_406_begin_0, end = var_406_end_0, end_mask = var_406_end_mask_0, squeeze_mask = var_406_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_406_cast_fp16")]; + bool var_407_transpose_x_0 = const()[name = string("op_407_transpose_x_0"), val = bool(false)]; + bool var_407_transpose_y_0 = const()[name = string("op_407_transpose_y_0"), val = bool(false)]; + tensor state_1_to_fp16 = const()[name = string("state_1_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66029184)))]; + tensor var_407_cast_fp16 = matmul(transpose_x = var_407_transpose_x_0, transpose_y = var_407_transpose_y_0, x = var_406_cast_fp16, y = state_1_to_fp16)[name = string("op_407_cast_fp16")]; + tensor v_new_1_cast_fp16 = sub(x = var_404_cast_fp16, y = var_407_cast_fp16)[name = string("v_new_1_cast_fp16")]; + tensor var_410_begin_0 = const()[name = string("op_410_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_410_end_0 = const()[name = string("op_410_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_410_end_mask_0 = const()[name = string("op_410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_410_squeeze_mask_0 = const()[name = string("op_410_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_410_cast_fp16 = slice_by_index(begin = var_410_begin_0, end = var_410_end_0, end_mask = var_410_end_mask_0, squeeze_mask = var_410_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_410_cast_fp16")]; + bool var_411_transpose_x_0 = const()[name = string("op_411_transpose_x_0"), val = bool(false)]; + bool var_411_transpose_y_0 = const()[name = string("op_411_transpose_y_0"), val = bool(false)]; + tensor var_411_cast_fp16 = matmul(transpose_x = var_411_transpose_x_0, transpose_y = var_411_transpose_y_0, x = var_410_cast_fp16, y = state_1_to_fp16)[name = string("op_411_cast_fp16")]; + tensor var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_413_squeeze_mask_0 = const()[name = string("op_413_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, squeeze_mask = var_413_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_413_cast_fp16")]; + bool var_414_transpose_x_0 = const()[name = string("op_414_transpose_x_0"), val = bool(false)]; + bool var_414_transpose_y_0 = const()[name = string("op_414_transpose_y_0"), val = bool(false)]; + tensor var_414_cast_fp16 = matmul(transpose_x = var_414_transpose_x_0, transpose_y = var_414_transpose_y_0, x = var_413_cast_fp16, y = v_new_1_cast_fp16)[name = string("op_414_cast_fp16")]; + tensor var_415_cast_fp16 = add(x = var_411_cast_fp16, y = var_414_cast_fp16)[name = string("op_415_cast_fp16")]; + tensor var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_417_end_0 = const()[name = string("op_417_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_417_squeeze_mask_0 = const()[name = string("op_417_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = var_417_end_0, end_mask = var_417_end_mask_0, squeeze_mask = var_417_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_417_cast_fp16")]; + tensor var_418_cast_fp16 = mul(x = state_1_to_fp16, y = var_417_cast_fp16)[name = string("op_418_cast_fp16")]; + tensor var_420_begin_0 = const()[name = string("op_420_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_420_end_0 = const()[name = string("op_420_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_420_end_mask_0 = const()[name = string("op_420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_420_squeeze_mask_0 = const()[name = string("op_420_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_420_cast_fp16 = slice_by_index(begin = var_420_begin_0, end = var_420_end_0, end_mask = var_420_end_mask_0, squeeze_mask = var_420_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_420_cast_fp16")]; + bool var_422_transpose_x_1 = const()[name = string("op_422_transpose_x_1"), val = bool(true)]; + bool var_422_transpose_y_1 = const()[name = string("op_422_transpose_y_1"), val = bool(false)]; + tensor var_422_cast_fp16 = matmul(transpose_x = var_422_transpose_x_1, transpose_y = var_422_transpose_y_1, x = var_420_cast_fp16, y = v_new_1_cast_fp16)[name = string("op_422_cast_fp16")]; + tensor state_3_cast_fp16 = add(x = var_418_cast_fp16, y = var_422_cast_fp16)[name = string("state_3_cast_fp16")]; + tensor var_425_begin_0 = const()[name = string("op_425_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_425_end_0 = const()[name = string("op_425_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_425_end_mask_0 = const()[name = string("op_425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_425_squeeze_mask_0 = const()[name = string("op_425_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_425_cast_fp16 = slice_by_index(begin = var_425_begin_0, end = var_425_end_0, end_mask = var_425_end_mask_0, squeeze_mask = var_425_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_425_cast_fp16")]; + tensor var_427_begin_0 = const()[name = string("op_427_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_427_end_0 = const()[name = string("op_427_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_427_end_mask_0 = const()[name = string("op_427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_427_squeeze_mask_0 = const()[name = string("op_427_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_427_cast_fp16 = slice_by_index(begin = var_427_begin_0, end = var_427_end_0, end_mask = var_427_end_mask_0, squeeze_mask = var_427_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_427_cast_fp16")]; + bool var_428_transpose_x_0 = const()[name = string("op_428_transpose_x_0"), val = bool(false)]; + bool var_428_transpose_y_0 = const()[name = string("op_428_transpose_y_0"), val = bool(false)]; + tensor var_428_cast_fp16 = matmul(transpose_x = var_428_transpose_x_0, transpose_y = var_428_transpose_y_0, x = var_427_cast_fp16, y = state_3_cast_fp16)[name = string("op_428_cast_fp16")]; + tensor v_new_3_cast_fp16 = sub(x = var_425_cast_fp16, y = var_428_cast_fp16)[name = string("v_new_3_cast_fp16")]; + tensor var_431_begin_0 = const()[name = string("op_431_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_431_end_0 = const()[name = string("op_431_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_431_end_mask_0 = const()[name = string("op_431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_431_squeeze_mask_0 = const()[name = string("op_431_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_431_cast_fp16 = slice_by_index(begin = var_431_begin_0, end = var_431_end_0, end_mask = var_431_end_mask_0, squeeze_mask = var_431_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_431_cast_fp16")]; + bool var_432_transpose_x_0 = const()[name = string("op_432_transpose_x_0"), val = bool(false)]; + bool var_432_transpose_y_0 = const()[name = string("op_432_transpose_y_0"), val = bool(false)]; + tensor var_432_cast_fp16 = matmul(transpose_x = var_432_transpose_x_0, transpose_y = var_432_transpose_y_0, x = var_431_cast_fp16, y = state_3_cast_fp16)[name = string("op_432_cast_fp16")]; + tensor var_434_begin_0 = const()[name = string("op_434_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_434_end_0 = const()[name = string("op_434_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_434_end_mask_0 = const()[name = string("op_434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_434_squeeze_mask_0 = const()[name = string("op_434_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_434_cast_fp16 = slice_by_index(begin = var_434_begin_0, end = var_434_end_0, end_mask = var_434_end_mask_0, squeeze_mask = var_434_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_434_cast_fp16")]; + bool var_435_transpose_x_0 = const()[name = string("op_435_transpose_x_0"), val = bool(false)]; + bool var_435_transpose_y_0 = const()[name = string("op_435_transpose_y_0"), val = bool(false)]; + tensor var_435_cast_fp16 = matmul(transpose_x = var_435_transpose_x_0, transpose_y = var_435_transpose_y_0, x = var_434_cast_fp16, y = v_new_3_cast_fp16)[name = string("op_435_cast_fp16")]; + tensor var_436_cast_fp16 = add(x = var_432_cast_fp16, y = var_435_cast_fp16)[name = string("op_436_cast_fp16")]; + tensor var_438_begin_0 = const()[name = string("op_438_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_438_end_0 = const()[name = string("op_438_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_438_end_mask_0 = const()[name = string("op_438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_438_squeeze_mask_0 = const()[name = string("op_438_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, squeeze_mask = var_438_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_438_cast_fp16")]; + tensor var_439_cast_fp16 = mul(x = state_3_cast_fp16, y = var_438_cast_fp16)[name = string("op_439_cast_fp16")]; + tensor var_441_begin_0 = const()[name = string("op_441_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_441_end_0 = const()[name = string("op_441_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_441_end_mask_0 = const()[name = string("op_441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_441_squeeze_mask_0 = const()[name = string("op_441_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_441_cast_fp16 = slice_by_index(begin = var_441_begin_0, end = var_441_end_0, end_mask = var_441_end_mask_0, squeeze_mask = var_441_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_441_cast_fp16")]; + bool var_443_transpose_x_1 = const()[name = string("op_443_transpose_x_1"), val = bool(true)]; + bool var_443_transpose_y_1 = const()[name = string("op_443_transpose_y_1"), val = bool(false)]; + tensor var_443_cast_fp16 = matmul(transpose_x = var_443_transpose_x_1, transpose_y = var_443_transpose_y_1, x = var_441_cast_fp16, y = v_new_3_cast_fp16)[name = string("op_443_cast_fp16")]; + tensor state_5_cast_fp16 = add(x = var_439_cast_fp16, y = var_443_cast_fp16)[name = string("state_5_cast_fp16")]; + tensor var_446_begin_0 = const()[name = string("op_446_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_446_end_0 = const()[name = string("op_446_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_446_end_mask_0 = const()[name = string("op_446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_446_squeeze_mask_0 = const()[name = string("op_446_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_446_cast_fp16 = slice_by_index(begin = var_446_begin_0, end = var_446_end_0, end_mask = var_446_end_mask_0, squeeze_mask = var_446_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_446_cast_fp16")]; + tensor var_448_begin_0 = const()[name = string("op_448_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_448_end_0 = const()[name = string("op_448_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_448_end_mask_0 = const()[name = string("op_448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_448_squeeze_mask_0 = const()[name = string("op_448_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_448_cast_fp16 = slice_by_index(begin = var_448_begin_0, end = var_448_end_0, end_mask = var_448_end_mask_0, squeeze_mask = var_448_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_448_cast_fp16")]; + bool var_449_transpose_x_0 = const()[name = string("op_449_transpose_x_0"), val = bool(false)]; + bool var_449_transpose_y_0 = const()[name = string("op_449_transpose_y_0"), val = bool(false)]; + tensor var_449_cast_fp16 = matmul(transpose_x = var_449_transpose_x_0, transpose_y = var_449_transpose_y_0, x = var_448_cast_fp16, y = state_5_cast_fp16)[name = string("op_449_cast_fp16")]; + tensor v_new_5_cast_fp16 = sub(x = var_446_cast_fp16, y = var_449_cast_fp16)[name = string("v_new_5_cast_fp16")]; + tensor var_452_begin_0 = const()[name = string("op_452_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_452_end_0 = const()[name = string("op_452_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_452_end_mask_0 = const()[name = string("op_452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_452_squeeze_mask_0 = const()[name = string("op_452_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_452_cast_fp16 = slice_by_index(begin = var_452_begin_0, end = var_452_end_0, end_mask = var_452_end_mask_0, squeeze_mask = var_452_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_452_cast_fp16")]; + bool var_453_transpose_x_0 = const()[name = string("op_453_transpose_x_0"), val = bool(false)]; + bool var_453_transpose_y_0 = const()[name = string("op_453_transpose_y_0"), val = bool(false)]; + tensor var_453_cast_fp16 = matmul(transpose_x = var_453_transpose_x_0, transpose_y = var_453_transpose_y_0, x = var_452_cast_fp16, y = state_5_cast_fp16)[name = string("op_453_cast_fp16")]; + tensor var_455_begin_0 = const()[name = string("op_455_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_455_end_0 = const()[name = string("op_455_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_455_end_mask_0 = const()[name = string("op_455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_455_squeeze_mask_0 = const()[name = string("op_455_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_455_cast_fp16 = slice_by_index(begin = var_455_begin_0, end = var_455_end_0, end_mask = var_455_end_mask_0, squeeze_mask = var_455_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_455_cast_fp16")]; + bool var_456_transpose_x_0 = const()[name = string("op_456_transpose_x_0"), val = bool(false)]; + bool var_456_transpose_y_0 = const()[name = string("op_456_transpose_y_0"), val = bool(false)]; + tensor var_456_cast_fp16 = matmul(transpose_x = var_456_transpose_x_0, transpose_y = var_456_transpose_y_0, x = var_455_cast_fp16, y = v_new_5_cast_fp16)[name = string("op_456_cast_fp16")]; + tensor var_457_cast_fp16 = add(x = var_453_cast_fp16, y = var_456_cast_fp16)[name = string("op_457_cast_fp16")]; + tensor var_459_begin_0 = const()[name = string("op_459_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_459_end_0 = const()[name = string("op_459_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_459_end_mask_0 = const()[name = string("op_459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_459_squeeze_mask_0 = const()[name = string("op_459_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_459_cast_fp16 = slice_by_index(begin = var_459_begin_0, end = var_459_end_0, end_mask = var_459_end_mask_0, squeeze_mask = var_459_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_459_cast_fp16")]; + tensor var_460_cast_fp16 = mul(x = state_5_cast_fp16, y = var_459_cast_fp16)[name = string("op_460_cast_fp16")]; + tensor var_462_begin_0 = const()[name = string("op_462_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_462_end_0 = const()[name = string("op_462_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_462_end_mask_0 = const()[name = string("op_462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_462_squeeze_mask_0 = const()[name = string("op_462_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_462_cast_fp16 = slice_by_index(begin = var_462_begin_0, end = var_462_end_0, end_mask = var_462_end_mask_0, squeeze_mask = var_462_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_462_cast_fp16")]; + bool var_464_transpose_x_1 = const()[name = string("op_464_transpose_x_1"), val = bool(true)]; + bool var_464_transpose_y_1 = const()[name = string("op_464_transpose_y_1"), val = bool(false)]; + tensor var_464_cast_fp16 = matmul(transpose_x = var_464_transpose_x_1, transpose_y = var_464_transpose_y_1, x = var_462_cast_fp16, y = v_new_5_cast_fp16)[name = string("op_464_cast_fp16")]; + tensor state_7_cast_fp16 = add(x = var_460_cast_fp16, y = var_464_cast_fp16)[name = string("state_7_cast_fp16")]; + tensor var_467_begin_0 = const()[name = string("op_467_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_467_end_0 = const()[name = string("op_467_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_467_end_mask_0 = const()[name = string("op_467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_467_squeeze_mask_0 = const()[name = string("op_467_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_467_cast_fp16 = slice_by_index(begin = var_467_begin_0, end = var_467_end_0, end_mask = var_467_end_mask_0, squeeze_mask = var_467_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_467_cast_fp16")]; + tensor var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_469_squeeze_mask_0 = const()[name = string("op_469_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, squeeze_mask = var_469_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_469_cast_fp16")]; + bool var_470_transpose_x_0 = const()[name = string("op_470_transpose_x_0"), val = bool(false)]; + bool var_470_transpose_y_0 = const()[name = string("op_470_transpose_y_0"), val = bool(false)]; + tensor var_470_cast_fp16 = matmul(transpose_x = var_470_transpose_x_0, transpose_y = var_470_transpose_y_0, x = var_469_cast_fp16, y = state_7_cast_fp16)[name = string("op_470_cast_fp16")]; + tensor v_new_7_cast_fp16 = sub(x = var_467_cast_fp16, y = var_470_cast_fp16)[name = string("v_new_7_cast_fp16")]; + tensor var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_473_squeeze_mask_0 = const()[name = string("op_473_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, squeeze_mask = var_473_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_473_cast_fp16")]; + bool var_474_transpose_x_0 = const()[name = string("op_474_transpose_x_0"), val = bool(false)]; + bool var_474_transpose_y_0 = const()[name = string("op_474_transpose_y_0"), val = bool(false)]; + tensor var_474_cast_fp16 = matmul(transpose_x = var_474_transpose_x_0, transpose_y = var_474_transpose_y_0, x = var_473_cast_fp16, y = state_7_cast_fp16)[name = string("op_474_cast_fp16")]; + tensor var_476_begin_0 = const()[name = string("op_476_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_476_end_0 = const()[name = string("op_476_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_476_end_mask_0 = const()[name = string("op_476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_476_squeeze_mask_0 = const()[name = string("op_476_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_476_cast_fp16 = slice_by_index(begin = var_476_begin_0, end = var_476_end_0, end_mask = var_476_end_mask_0, squeeze_mask = var_476_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_476_cast_fp16")]; + bool var_477_transpose_x_0 = const()[name = string("op_477_transpose_x_0"), val = bool(false)]; + bool var_477_transpose_y_0 = const()[name = string("op_477_transpose_y_0"), val = bool(false)]; + tensor var_477_cast_fp16 = matmul(transpose_x = var_477_transpose_x_0, transpose_y = var_477_transpose_y_0, x = var_476_cast_fp16, y = v_new_7_cast_fp16)[name = string("op_477_cast_fp16")]; + tensor var_478_cast_fp16 = add(x = var_474_cast_fp16, y = var_477_cast_fp16)[name = string("op_478_cast_fp16")]; + tensor var_480_begin_0 = const()[name = string("op_480_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_480_end_0 = const()[name = string("op_480_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_480_end_mask_0 = const()[name = string("op_480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_480_squeeze_mask_0 = const()[name = string("op_480_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_480_cast_fp16 = slice_by_index(begin = var_480_begin_0, end = var_480_end_0, end_mask = var_480_end_mask_0, squeeze_mask = var_480_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_480_cast_fp16")]; + tensor var_481_cast_fp16 = mul(x = state_7_cast_fp16, y = var_480_cast_fp16)[name = string("op_481_cast_fp16")]; + tensor var_483_begin_0 = const()[name = string("op_483_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_483_end_0 = const()[name = string("op_483_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_483_end_mask_0 = const()[name = string("op_483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_483_squeeze_mask_0 = const()[name = string("op_483_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_483_cast_fp16 = slice_by_index(begin = var_483_begin_0, end = var_483_end_0, end_mask = var_483_end_mask_0, squeeze_mask = var_483_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_483_cast_fp16")]; + bool var_485_transpose_x_1 = const()[name = string("op_485_transpose_x_1"), val = bool(true)]; + bool var_485_transpose_y_1 = const()[name = string("op_485_transpose_y_1"), val = bool(false)]; + tensor var_485_cast_fp16 = matmul(transpose_x = var_485_transpose_x_1, transpose_y = var_485_transpose_y_1, x = var_483_cast_fp16, y = v_new_7_cast_fp16)[name = string("op_485_cast_fp16")]; + tensor state_9_cast_fp16 = add(x = var_481_cast_fp16, y = var_485_cast_fp16)[name = string("state_9_cast_fp16")]; + tensor var_488_begin_0 = const()[name = string("op_488_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_488_end_0 = const()[name = string("op_488_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_488_end_mask_0 = const()[name = string("op_488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_488_squeeze_mask_0 = const()[name = string("op_488_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_488_cast_fp16 = slice_by_index(begin = var_488_begin_0, end = var_488_end_0, end_mask = var_488_end_mask_0, squeeze_mask = var_488_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_488_cast_fp16")]; + tensor var_490_begin_0 = const()[name = string("op_490_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_490_end_0 = const()[name = string("op_490_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_490_end_mask_0 = const()[name = string("op_490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_490_squeeze_mask_0 = const()[name = string("op_490_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_490_cast_fp16 = slice_by_index(begin = var_490_begin_0, end = var_490_end_0, end_mask = var_490_end_mask_0, squeeze_mask = var_490_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_490_cast_fp16")]; + bool var_491_transpose_x_0 = const()[name = string("op_491_transpose_x_0"), val = bool(false)]; + bool var_491_transpose_y_0 = const()[name = string("op_491_transpose_y_0"), val = bool(false)]; + tensor var_491_cast_fp16 = matmul(transpose_x = var_491_transpose_x_0, transpose_y = var_491_transpose_y_0, x = var_490_cast_fp16, y = state_9_cast_fp16)[name = string("op_491_cast_fp16")]; + tensor v_new_9_cast_fp16 = sub(x = var_488_cast_fp16, y = var_491_cast_fp16)[name = string("v_new_9_cast_fp16")]; + tensor var_494_begin_0 = const()[name = string("op_494_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_494_end_0 = const()[name = string("op_494_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_494_end_mask_0 = const()[name = string("op_494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_494_squeeze_mask_0 = const()[name = string("op_494_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_494_cast_fp16 = slice_by_index(begin = var_494_begin_0, end = var_494_end_0, end_mask = var_494_end_mask_0, squeeze_mask = var_494_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_494_cast_fp16")]; + bool var_495_transpose_x_0 = const()[name = string("op_495_transpose_x_0"), val = bool(false)]; + bool var_495_transpose_y_0 = const()[name = string("op_495_transpose_y_0"), val = bool(false)]; + tensor var_495_cast_fp16 = matmul(transpose_x = var_495_transpose_x_0, transpose_y = var_495_transpose_y_0, x = var_494_cast_fp16, y = state_9_cast_fp16)[name = string("op_495_cast_fp16")]; + tensor var_497_begin_0 = const()[name = string("op_497_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_497_end_0 = const()[name = string("op_497_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_497_end_mask_0 = const()[name = string("op_497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_497_squeeze_mask_0 = const()[name = string("op_497_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_497_cast_fp16 = slice_by_index(begin = var_497_begin_0, end = var_497_end_0, end_mask = var_497_end_mask_0, squeeze_mask = var_497_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_497_cast_fp16")]; + bool var_498_transpose_x_0 = const()[name = string("op_498_transpose_x_0"), val = bool(false)]; + bool var_498_transpose_y_0 = const()[name = string("op_498_transpose_y_0"), val = bool(false)]; + tensor var_498_cast_fp16 = matmul(transpose_x = var_498_transpose_x_0, transpose_y = var_498_transpose_y_0, x = var_497_cast_fp16, y = v_new_9_cast_fp16)[name = string("op_498_cast_fp16")]; + tensor var_499_cast_fp16 = add(x = var_495_cast_fp16, y = var_498_cast_fp16)[name = string("op_499_cast_fp16")]; + tensor var_501_begin_0 = const()[name = string("op_501_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_501_end_0 = const()[name = string("op_501_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_501_end_mask_0 = const()[name = string("op_501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_501_squeeze_mask_0 = const()[name = string("op_501_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_501_cast_fp16 = slice_by_index(begin = var_501_begin_0, end = var_501_end_0, end_mask = var_501_end_mask_0, squeeze_mask = var_501_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_501_cast_fp16")]; + tensor var_502_cast_fp16 = mul(x = state_9_cast_fp16, y = var_501_cast_fp16)[name = string("op_502_cast_fp16")]; + tensor var_504_begin_0 = const()[name = string("op_504_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_504_end_0 = const()[name = string("op_504_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_504_end_mask_0 = const()[name = string("op_504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_504_squeeze_mask_0 = const()[name = string("op_504_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_504_cast_fp16 = slice_by_index(begin = var_504_begin_0, end = var_504_end_0, end_mask = var_504_end_mask_0, squeeze_mask = var_504_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_504_cast_fp16")]; + bool var_506_transpose_x_1 = const()[name = string("op_506_transpose_x_1"), val = bool(true)]; + bool var_506_transpose_y_1 = const()[name = string("op_506_transpose_y_1"), val = bool(false)]; + tensor var_506_cast_fp16 = matmul(transpose_x = var_506_transpose_x_1, transpose_y = var_506_transpose_y_1, x = var_504_cast_fp16, y = v_new_9_cast_fp16)[name = string("op_506_cast_fp16")]; + tensor state_11_cast_fp16 = add(x = var_502_cast_fp16, y = var_506_cast_fp16)[name = string("state_11_cast_fp16")]; + tensor var_509_begin_0 = const()[name = string("op_509_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_509_end_0 = const()[name = string("op_509_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_509_end_mask_0 = const()[name = string("op_509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_509_squeeze_mask_0 = const()[name = string("op_509_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_509_cast_fp16 = slice_by_index(begin = var_509_begin_0, end = var_509_end_0, end_mask = var_509_end_mask_0, squeeze_mask = var_509_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_509_cast_fp16")]; + tensor var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_squeeze_mask_0 = const()[name = string("op_511_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, squeeze_mask = var_511_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_511_cast_fp16")]; + bool var_512_transpose_x_0 = const()[name = string("op_512_transpose_x_0"), val = bool(false)]; + bool var_512_transpose_y_0 = const()[name = string("op_512_transpose_y_0"), val = bool(false)]; + tensor var_512_cast_fp16 = matmul(transpose_x = var_512_transpose_x_0, transpose_y = var_512_transpose_y_0, x = var_511_cast_fp16, y = state_11_cast_fp16)[name = string("op_512_cast_fp16")]; + tensor v_new_11_cast_fp16 = sub(x = var_509_cast_fp16, y = var_512_cast_fp16)[name = string("v_new_11_cast_fp16")]; + tensor var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_squeeze_mask_0 = const()[name = string("op_515_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, squeeze_mask = var_515_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_515_cast_fp16")]; + bool var_516_transpose_x_0 = const()[name = string("op_516_transpose_x_0"), val = bool(false)]; + bool var_516_transpose_y_0 = const()[name = string("op_516_transpose_y_0"), val = bool(false)]; + tensor var_516_cast_fp16 = matmul(transpose_x = var_516_transpose_x_0, transpose_y = var_516_transpose_y_0, x = var_515_cast_fp16, y = state_11_cast_fp16)[name = string("op_516_cast_fp16")]; + tensor var_518_begin_0 = const()[name = string("op_518_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_518_end_0 = const()[name = string("op_518_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_518_end_mask_0 = const()[name = string("op_518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_518_squeeze_mask_0 = const()[name = string("op_518_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_518_cast_fp16 = slice_by_index(begin = var_518_begin_0, end = var_518_end_0, end_mask = var_518_end_mask_0, squeeze_mask = var_518_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_518_cast_fp16")]; + bool var_519_transpose_x_0 = const()[name = string("op_519_transpose_x_0"), val = bool(false)]; + bool var_519_transpose_y_0 = const()[name = string("op_519_transpose_y_0"), val = bool(false)]; + tensor var_519_cast_fp16 = matmul(transpose_x = var_519_transpose_x_0, transpose_y = var_519_transpose_y_0, x = var_518_cast_fp16, y = v_new_11_cast_fp16)[name = string("op_519_cast_fp16")]; + tensor var_520_cast_fp16 = add(x = var_516_cast_fp16, y = var_519_cast_fp16)[name = string("op_520_cast_fp16")]; + tensor var_522_begin_0 = const()[name = string("op_522_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_522_end_0 = const()[name = string("op_522_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_522_end_mask_0 = const()[name = string("op_522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_522_squeeze_mask_0 = const()[name = string("op_522_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_522_cast_fp16 = slice_by_index(begin = var_522_begin_0, end = var_522_end_0, end_mask = var_522_end_mask_0, squeeze_mask = var_522_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_522_cast_fp16")]; + tensor var_523_cast_fp16 = mul(x = state_11_cast_fp16, y = var_522_cast_fp16)[name = string("op_523_cast_fp16")]; + tensor var_525_begin_0 = const()[name = string("op_525_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_525_end_0 = const()[name = string("op_525_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_525_end_mask_0 = const()[name = string("op_525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_525_squeeze_mask_0 = const()[name = string("op_525_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_525_cast_fp16 = slice_by_index(begin = var_525_begin_0, end = var_525_end_0, end_mask = var_525_end_mask_0, squeeze_mask = var_525_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_525_cast_fp16")]; + bool var_527_transpose_x_1 = const()[name = string("op_527_transpose_x_1"), val = bool(true)]; + bool var_527_transpose_y_1 = const()[name = string("op_527_transpose_y_1"), val = bool(false)]; + tensor var_527_cast_fp16 = matmul(transpose_x = var_527_transpose_x_1, transpose_y = var_527_transpose_y_1, x = var_525_cast_fp16, y = v_new_11_cast_fp16)[name = string("op_527_cast_fp16")]; + tensor state_13_cast_fp16 = add(x = var_523_cast_fp16, y = var_527_cast_fp16)[name = string("state_13_cast_fp16")]; + tensor var_530_begin_0 = const()[name = string("op_530_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_530_end_0 = const()[name = string("op_530_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_530_end_mask_0 = const()[name = string("op_530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_530_squeeze_mask_0 = const()[name = string("op_530_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_530_cast_fp16 = slice_by_index(begin = var_530_begin_0, end = var_530_end_0, end_mask = var_530_end_mask_0, squeeze_mask = var_530_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_530_cast_fp16")]; + tensor var_532_begin_0 = const()[name = string("op_532_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_532_end_0 = const()[name = string("op_532_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_532_end_mask_0 = const()[name = string("op_532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_532_squeeze_mask_0 = const()[name = string("op_532_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_532_cast_fp16 = slice_by_index(begin = var_532_begin_0, end = var_532_end_0, end_mask = var_532_end_mask_0, squeeze_mask = var_532_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_532_cast_fp16")]; + bool var_533_transpose_x_0 = const()[name = string("op_533_transpose_x_0"), val = bool(false)]; + bool var_533_transpose_y_0 = const()[name = string("op_533_transpose_y_0"), val = bool(false)]; + tensor var_533_cast_fp16 = matmul(transpose_x = var_533_transpose_x_0, transpose_y = var_533_transpose_y_0, x = var_532_cast_fp16, y = state_13_cast_fp16)[name = string("op_533_cast_fp16")]; + tensor v_new_13_cast_fp16 = sub(x = var_530_cast_fp16, y = var_533_cast_fp16)[name = string("v_new_13_cast_fp16")]; + tensor var_536_begin_0 = const()[name = string("op_536_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_536_end_0 = const()[name = string("op_536_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_536_end_mask_0 = const()[name = string("op_536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_536_squeeze_mask_0 = const()[name = string("op_536_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_536_cast_fp16 = slice_by_index(begin = var_536_begin_0, end = var_536_end_0, end_mask = var_536_end_mask_0, squeeze_mask = var_536_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_536_cast_fp16")]; + bool var_537_transpose_x_0 = const()[name = string("op_537_transpose_x_0"), val = bool(false)]; + bool var_537_transpose_y_0 = const()[name = string("op_537_transpose_y_0"), val = bool(false)]; + tensor var_537_cast_fp16 = matmul(transpose_x = var_537_transpose_x_0, transpose_y = var_537_transpose_y_0, x = var_536_cast_fp16, y = state_13_cast_fp16)[name = string("op_537_cast_fp16")]; + tensor var_539_begin_0 = const()[name = string("op_539_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_539_end_0 = const()[name = string("op_539_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_539_end_mask_0 = const()[name = string("op_539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_539_squeeze_mask_0 = const()[name = string("op_539_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_539_cast_fp16 = slice_by_index(begin = var_539_begin_0, end = var_539_end_0, end_mask = var_539_end_mask_0, squeeze_mask = var_539_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_539_cast_fp16")]; + bool var_540_transpose_x_0 = const()[name = string("op_540_transpose_x_0"), val = bool(false)]; + bool var_540_transpose_y_0 = const()[name = string("op_540_transpose_y_0"), val = bool(false)]; + tensor var_540_cast_fp16 = matmul(transpose_x = var_540_transpose_x_0, transpose_y = var_540_transpose_y_0, x = var_539_cast_fp16, y = v_new_13_cast_fp16)[name = string("op_540_cast_fp16")]; + tensor var_541_cast_fp16 = add(x = var_537_cast_fp16, y = var_540_cast_fp16)[name = string("op_541_cast_fp16")]; + tensor var_543_begin_0 = const()[name = string("op_543_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_543_end_0 = const()[name = string("op_543_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_543_end_mask_0 = const()[name = string("op_543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_543_squeeze_mask_0 = const()[name = string("op_543_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_543_cast_fp16 = slice_by_index(begin = var_543_begin_0, end = var_543_end_0, end_mask = var_543_end_mask_0, squeeze_mask = var_543_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_543_cast_fp16")]; + tensor var_544_cast_fp16 = mul(x = state_13_cast_fp16, y = var_543_cast_fp16)[name = string("op_544_cast_fp16")]; + tensor var_546_begin_0 = const()[name = string("op_546_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_546_end_0 = const()[name = string("op_546_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_546_end_mask_0 = const()[name = string("op_546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_546_squeeze_mask_0 = const()[name = string("op_546_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_546_cast_fp16 = slice_by_index(begin = var_546_begin_0, end = var_546_end_0, end_mask = var_546_end_mask_0, squeeze_mask = var_546_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_546_cast_fp16")]; + bool var_548_transpose_x_1 = const()[name = string("op_548_transpose_x_1"), val = bool(true)]; + bool var_548_transpose_y_1 = const()[name = string("op_548_transpose_y_1"), val = bool(false)]; + tensor var_548_cast_fp16 = matmul(transpose_x = var_548_transpose_x_1, transpose_y = var_548_transpose_y_1, x = var_546_cast_fp16, y = v_new_13_cast_fp16)[name = string("op_548_cast_fp16")]; + tensor state_15_cast_fp16 = add(x = var_544_cast_fp16, y = var_548_cast_fp16)[name = string("state_15_cast_fp16")]; + tensor var_551_begin_0 = const()[name = string("op_551_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_551_end_0 = const()[name = string("op_551_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_551_end_mask_0 = const()[name = string("op_551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_551_squeeze_mask_0 = const()[name = string("op_551_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_551_cast_fp16 = slice_by_index(begin = var_551_begin_0, end = var_551_end_0, end_mask = var_551_end_mask_0, squeeze_mask = var_551_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_551_cast_fp16")]; + tensor var_553_begin_0 = const()[name = string("op_553_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_553_end_0 = const()[name = string("op_553_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_553_end_mask_0 = const()[name = string("op_553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_553_squeeze_mask_0 = const()[name = string("op_553_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_553_cast_fp16 = slice_by_index(begin = var_553_begin_0, end = var_553_end_0, end_mask = var_553_end_mask_0, squeeze_mask = var_553_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_553_cast_fp16")]; + bool var_554_transpose_x_0 = const()[name = string("op_554_transpose_x_0"), val = bool(false)]; + bool var_554_transpose_y_0 = const()[name = string("op_554_transpose_y_0"), val = bool(false)]; + tensor var_554_cast_fp16 = matmul(transpose_x = var_554_transpose_x_0, transpose_y = var_554_transpose_y_0, x = var_553_cast_fp16, y = state_15_cast_fp16)[name = string("op_554_cast_fp16")]; + tensor v_new_15_cast_fp16 = sub(x = var_551_cast_fp16, y = var_554_cast_fp16)[name = string("v_new_15_cast_fp16")]; + tensor var_557_begin_0 = const()[name = string("op_557_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_557_end_0 = const()[name = string("op_557_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_557_end_mask_0 = const()[name = string("op_557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_557_squeeze_mask_0 = const()[name = string("op_557_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_557_cast_fp16 = slice_by_index(begin = var_557_begin_0, end = var_557_end_0, end_mask = var_557_end_mask_0, squeeze_mask = var_557_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_557_cast_fp16")]; + bool var_558_transpose_x_0 = const()[name = string("op_558_transpose_x_0"), val = bool(false)]; + bool var_558_transpose_y_0 = const()[name = string("op_558_transpose_y_0"), val = bool(false)]; + tensor var_558_cast_fp16 = matmul(transpose_x = var_558_transpose_x_0, transpose_y = var_558_transpose_y_0, x = var_557_cast_fp16, y = state_15_cast_fp16)[name = string("op_558_cast_fp16")]; + tensor var_560_begin_0 = const()[name = string("op_560_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_560_end_0 = const()[name = string("op_560_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_560_end_mask_0 = const()[name = string("op_560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_560_squeeze_mask_0 = const()[name = string("op_560_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_560_cast_fp16 = slice_by_index(begin = var_560_begin_0, end = var_560_end_0, end_mask = var_560_end_mask_0, squeeze_mask = var_560_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_560_cast_fp16")]; + bool var_561_transpose_x_0 = const()[name = string("op_561_transpose_x_0"), val = bool(false)]; + bool var_561_transpose_y_0 = const()[name = string("op_561_transpose_y_0"), val = bool(false)]; + tensor var_561_cast_fp16 = matmul(transpose_x = var_561_transpose_x_0, transpose_y = var_561_transpose_y_0, x = var_560_cast_fp16, y = v_new_15_cast_fp16)[name = string("op_561_cast_fp16")]; + tensor var_562_cast_fp16 = add(x = var_558_cast_fp16, y = var_561_cast_fp16)[name = string("op_562_cast_fp16")]; + tensor var_564_begin_0 = const()[name = string("op_564_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_564_end_0 = const()[name = string("op_564_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_564_end_mask_0 = const()[name = string("op_564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_564_squeeze_mask_0 = const()[name = string("op_564_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_564_cast_fp16 = slice_by_index(begin = var_564_begin_0, end = var_564_end_0, end_mask = var_564_end_mask_0, squeeze_mask = var_564_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_564_cast_fp16")]; + tensor var_565_cast_fp16 = mul(x = state_15_cast_fp16, y = var_564_cast_fp16)[name = string("op_565_cast_fp16")]; + tensor var_567_begin_0 = const()[name = string("op_567_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_567_end_0 = const()[name = string("op_567_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_567_end_mask_0 = const()[name = string("op_567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_567_squeeze_mask_0 = const()[name = string("op_567_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_567_cast_fp16 = slice_by_index(begin = var_567_begin_0, end = var_567_end_0, end_mask = var_567_end_mask_0, squeeze_mask = var_567_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_567_cast_fp16")]; + bool var_569_transpose_x_1 = const()[name = string("op_569_transpose_x_1"), val = bool(true)]; + bool var_569_transpose_y_1 = const()[name = string("op_569_transpose_y_1"), val = bool(false)]; + tensor var_569_cast_fp16 = matmul(transpose_x = var_569_transpose_x_1, transpose_y = var_569_transpose_y_1, x = var_567_cast_fp16, y = v_new_15_cast_fp16)[name = string("op_569_cast_fp16")]; + tensor state_17_cast_fp16 = add(x = var_565_cast_fp16, y = var_569_cast_fp16)[name = string("state_17_cast_fp16")]; + tensor var_572_begin_0 = const()[name = string("op_572_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_572_end_0 = const()[name = string("op_572_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_572_end_mask_0 = const()[name = string("op_572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_572_squeeze_mask_0 = const()[name = string("op_572_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_572_cast_fp16 = slice_by_index(begin = var_572_begin_0, end = var_572_end_0, end_mask = var_572_end_mask_0, squeeze_mask = var_572_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_572_cast_fp16")]; + tensor var_574_begin_0 = const()[name = string("op_574_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_574_end_0 = const()[name = string("op_574_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_574_end_mask_0 = const()[name = string("op_574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_574_squeeze_mask_0 = const()[name = string("op_574_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_574_cast_fp16 = slice_by_index(begin = var_574_begin_0, end = var_574_end_0, end_mask = var_574_end_mask_0, squeeze_mask = var_574_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_574_cast_fp16")]; + bool var_575_transpose_x_0 = const()[name = string("op_575_transpose_x_0"), val = bool(false)]; + bool var_575_transpose_y_0 = const()[name = string("op_575_transpose_y_0"), val = bool(false)]; + tensor var_575_cast_fp16 = matmul(transpose_x = var_575_transpose_x_0, transpose_y = var_575_transpose_y_0, x = var_574_cast_fp16, y = state_17_cast_fp16)[name = string("op_575_cast_fp16")]; + tensor v_new_17_cast_fp16 = sub(x = var_572_cast_fp16, y = var_575_cast_fp16)[name = string("v_new_17_cast_fp16")]; + tensor var_578_begin_0 = const()[name = string("op_578_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_578_end_0 = const()[name = string("op_578_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_578_end_mask_0 = const()[name = string("op_578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_578_squeeze_mask_0 = const()[name = string("op_578_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_578_cast_fp16 = slice_by_index(begin = var_578_begin_0, end = var_578_end_0, end_mask = var_578_end_mask_0, squeeze_mask = var_578_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_578_cast_fp16")]; + bool var_579_transpose_x_0 = const()[name = string("op_579_transpose_x_0"), val = bool(false)]; + bool var_579_transpose_y_0 = const()[name = string("op_579_transpose_y_0"), val = bool(false)]; + tensor var_579_cast_fp16 = matmul(transpose_x = var_579_transpose_x_0, transpose_y = var_579_transpose_y_0, x = var_578_cast_fp16, y = state_17_cast_fp16)[name = string("op_579_cast_fp16")]; + tensor var_581_begin_0 = const()[name = string("op_581_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_581_end_0 = const()[name = string("op_581_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_581_end_mask_0 = const()[name = string("op_581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_581_squeeze_mask_0 = const()[name = string("op_581_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_581_cast_fp16 = slice_by_index(begin = var_581_begin_0, end = var_581_end_0, end_mask = var_581_end_mask_0, squeeze_mask = var_581_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_581_cast_fp16")]; + bool var_582_transpose_x_0 = const()[name = string("op_582_transpose_x_0"), val = bool(false)]; + bool var_582_transpose_y_0 = const()[name = string("op_582_transpose_y_0"), val = bool(false)]; + tensor var_582_cast_fp16 = matmul(transpose_x = var_582_transpose_x_0, transpose_y = var_582_transpose_y_0, x = var_581_cast_fp16, y = v_new_17_cast_fp16)[name = string("op_582_cast_fp16")]; + tensor var_583_cast_fp16 = add(x = var_579_cast_fp16, y = var_582_cast_fp16)[name = string("op_583_cast_fp16")]; + tensor var_585_begin_0 = const()[name = string("op_585_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_585_end_0 = const()[name = string("op_585_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_585_end_mask_0 = const()[name = string("op_585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_585_squeeze_mask_0 = const()[name = string("op_585_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_585_cast_fp16 = slice_by_index(begin = var_585_begin_0, end = var_585_end_0, end_mask = var_585_end_mask_0, squeeze_mask = var_585_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_585_cast_fp16")]; + tensor var_586_cast_fp16 = mul(x = state_17_cast_fp16, y = var_585_cast_fp16)[name = string("op_586_cast_fp16")]; + tensor var_588_begin_0 = const()[name = string("op_588_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_588_end_0 = const()[name = string("op_588_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_588_end_mask_0 = const()[name = string("op_588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_588_squeeze_mask_0 = const()[name = string("op_588_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_588_cast_fp16 = slice_by_index(begin = var_588_begin_0, end = var_588_end_0, end_mask = var_588_end_mask_0, squeeze_mask = var_588_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_588_cast_fp16")]; + bool var_590_transpose_x_1 = const()[name = string("op_590_transpose_x_1"), val = bool(true)]; + bool var_590_transpose_y_1 = const()[name = string("op_590_transpose_y_1"), val = bool(false)]; + tensor var_590_cast_fp16 = matmul(transpose_x = var_590_transpose_x_1, transpose_y = var_590_transpose_y_1, x = var_588_cast_fp16, y = v_new_17_cast_fp16)[name = string("op_590_cast_fp16")]; + tensor state_19_cast_fp16 = add(x = var_586_cast_fp16, y = var_590_cast_fp16)[name = string("state_19_cast_fp16")]; + tensor var_593_begin_0 = const()[name = string("op_593_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_593_end_0 = const()[name = string("op_593_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_593_end_mask_0 = const()[name = string("op_593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_593_squeeze_mask_0 = const()[name = string("op_593_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_593_cast_fp16 = slice_by_index(begin = var_593_begin_0, end = var_593_end_0, end_mask = var_593_end_mask_0, squeeze_mask = var_593_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_593_cast_fp16")]; + tensor var_595_begin_0 = const()[name = string("op_595_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_595_end_0 = const()[name = string("op_595_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_595_end_mask_0 = const()[name = string("op_595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_595_squeeze_mask_0 = const()[name = string("op_595_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_595_cast_fp16 = slice_by_index(begin = var_595_begin_0, end = var_595_end_0, end_mask = var_595_end_mask_0, squeeze_mask = var_595_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_595_cast_fp16")]; + bool var_596_transpose_x_0 = const()[name = string("op_596_transpose_x_0"), val = bool(false)]; + bool var_596_transpose_y_0 = const()[name = string("op_596_transpose_y_0"), val = bool(false)]; + tensor var_596_cast_fp16 = matmul(transpose_x = var_596_transpose_x_0, transpose_y = var_596_transpose_y_0, x = var_595_cast_fp16, y = state_19_cast_fp16)[name = string("op_596_cast_fp16")]; + tensor v_new_19_cast_fp16 = sub(x = var_593_cast_fp16, y = var_596_cast_fp16)[name = string("v_new_19_cast_fp16")]; + tensor var_599_begin_0 = const()[name = string("op_599_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_599_end_0 = const()[name = string("op_599_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_599_end_mask_0 = const()[name = string("op_599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_599_squeeze_mask_0 = const()[name = string("op_599_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_599_cast_fp16 = slice_by_index(begin = var_599_begin_0, end = var_599_end_0, end_mask = var_599_end_mask_0, squeeze_mask = var_599_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_599_cast_fp16")]; + bool var_600_transpose_x_0 = const()[name = string("op_600_transpose_x_0"), val = bool(false)]; + bool var_600_transpose_y_0 = const()[name = string("op_600_transpose_y_0"), val = bool(false)]; + tensor var_600_cast_fp16 = matmul(transpose_x = var_600_transpose_x_0, transpose_y = var_600_transpose_y_0, x = var_599_cast_fp16, y = state_19_cast_fp16)[name = string("op_600_cast_fp16")]; + tensor var_602_begin_0 = const()[name = string("op_602_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_602_end_0 = const()[name = string("op_602_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_602_end_mask_0 = const()[name = string("op_602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_602_squeeze_mask_0 = const()[name = string("op_602_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_602_cast_fp16 = slice_by_index(begin = var_602_begin_0, end = var_602_end_0, end_mask = var_602_end_mask_0, squeeze_mask = var_602_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_602_cast_fp16")]; + bool var_603_transpose_x_0 = const()[name = string("op_603_transpose_x_0"), val = bool(false)]; + bool var_603_transpose_y_0 = const()[name = string("op_603_transpose_y_0"), val = bool(false)]; + tensor var_603_cast_fp16 = matmul(transpose_x = var_603_transpose_x_0, transpose_y = var_603_transpose_y_0, x = var_602_cast_fp16, y = v_new_19_cast_fp16)[name = string("op_603_cast_fp16")]; + tensor var_604_cast_fp16 = add(x = var_600_cast_fp16, y = var_603_cast_fp16)[name = string("op_604_cast_fp16")]; + tensor var_606_begin_0 = const()[name = string("op_606_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_606_end_0 = const()[name = string("op_606_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_606_end_mask_0 = const()[name = string("op_606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_606_squeeze_mask_0 = const()[name = string("op_606_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_606_cast_fp16 = slice_by_index(begin = var_606_begin_0, end = var_606_end_0, end_mask = var_606_end_mask_0, squeeze_mask = var_606_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_606_cast_fp16")]; + tensor var_607_cast_fp16 = mul(x = state_19_cast_fp16, y = var_606_cast_fp16)[name = string("op_607_cast_fp16")]; + tensor var_609_begin_0 = const()[name = string("op_609_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_609_end_0 = const()[name = string("op_609_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_609_end_mask_0 = const()[name = string("op_609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_609_squeeze_mask_0 = const()[name = string("op_609_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_609_cast_fp16 = slice_by_index(begin = var_609_begin_0, end = var_609_end_0, end_mask = var_609_end_mask_0, squeeze_mask = var_609_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_609_cast_fp16")]; + bool var_611_transpose_x_1 = const()[name = string("op_611_transpose_x_1"), val = bool(true)]; + bool var_611_transpose_y_1 = const()[name = string("op_611_transpose_y_1"), val = bool(false)]; + tensor var_611_cast_fp16 = matmul(transpose_x = var_611_transpose_x_1, transpose_y = var_611_transpose_y_1, x = var_609_cast_fp16, y = v_new_19_cast_fp16)[name = string("op_611_cast_fp16")]; + tensor state_21_cast_fp16 = add(x = var_607_cast_fp16, y = var_611_cast_fp16)[name = string("state_21_cast_fp16")]; + tensor var_614_begin_0 = const()[name = string("op_614_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_614_end_0 = const()[name = string("op_614_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_614_end_mask_0 = const()[name = string("op_614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_614_squeeze_mask_0 = const()[name = string("op_614_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_614_cast_fp16 = slice_by_index(begin = var_614_begin_0, end = var_614_end_0, end_mask = var_614_end_mask_0, squeeze_mask = var_614_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_614_cast_fp16")]; + tensor var_616_begin_0 = const()[name = string("op_616_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_616_end_0 = const()[name = string("op_616_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_616_end_mask_0 = const()[name = string("op_616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_616_squeeze_mask_0 = const()[name = string("op_616_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_616_cast_fp16 = slice_by_index(begin = var_616_begin_0, end = var_616_end_0, end_mask = var_616_end_mask_0, squeeze_mask = var_616_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_616_cast_fp16")]; + bool var_617_transpose_x_0 = const()[name = string("op_617_transpose_x_0"), val = bool(false)]; + bool var_617_transpose_y_0 = const()[name = string("op_617_transpose_y_0"), val = bool(false)]; + tensor var_617_cast_fp16 = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_616_cast_fp16, y = state_21_cast_fp16)[name = string("op_617_cast_fp16")]; + tensor v_new_21_cast_fp16 = sub(x = var_614_cast_fp16, y = var_617_cast_fp16)[name = string("v_new_21_cast_fp16")]; + tensor var_620_begin_0 = const()[name = string("op_620_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_620_end_0 = const()[name = string("op_620_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_620_end_mask_0 = const()[name = string("op_620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_620_squeeze_mask_0 = const()[name = string("op_620_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_620_cast_fp16 = slice_by_index(begin = var_620_begin_0, end = var_620_end_0, end_mask = var_620_end_mask_0, squeeze_mask = var_620_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_620_cast_fp16")]; + bool var_621_transpose_x_0 = const()[name = string("op_621_transpose_x_0"), val = bool(false)]; + bool var_621_transpose_y_0 = const()[name = string("op_621_transpose_y_0"), val = bool(false)]; + tensor var_621_cast_fp16 = matmul(transpose_x = var_621_transpose_x_0, transpose_y = var_621_transpose_y_0, x = var_620_cast_fp16, y = state_21_cast_fp16)[name = string("op_621_cast_fp16")]; + tensor var_623_begin_0 = const()[name = string("op_623_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_623_end_0 = const()[name = string("op_623_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_623_end_mask_0 = const()[name = string("op_623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_623_squeeze_mask_0 = const()[name = string("op_623_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_623_cast_fp16 = slice_by_index(begin = var_623_begin_0, end = var_623_end_0, end_mask = var_623_end_mask_0, squeeze_mask = var_623_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_623_cast_fp16")]; + bool var_624_transpose_x_0 = const()[name = string("op_624_transpose_x_0"), val = bool(false)]; + bool var_624_transpose_y_0 = const()[name = string("op_624_transpose_y_0"), val = bool(false)]; + tensor var_624_cast_fp16 = matmul(transpose_x = var_624_transpose_x_0, transpose_y = var_624_transpose_y_0, x = var_623_cast_fp16, y = v_new_21_cast_fp16)[name = string("op_624_cast_fp16")]; + tensor var_625_cast_fp16 = add(x = var_621_cast_fp16, y = var_624_cast_fp16)[name = string("op_625_cast_fp16")]; + tensor var_627_begin_0 = const()[name = string("op_627_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_627_end_0 = const()[name = string("op_627_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_627_end_mask_0 = const()[name = string("op_627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_627_squeeze_mask_0 = const()[name = string("op_627_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_627_cast_fp16 = slice_by_index(begin = var_627_begin_0, end = var_627_end_0, end_mask = var_627_end_mask_0, squeeze_mask = var_627_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_627_cast_fp16")]; + tensor var_628_cast_fp16 = mul(x = state_21_cast_fp16, y = var_627_cast_fp16)[name = string("op_628_cast_fp16")]; + tensor var_630_begin_0 = const()[name = string("op_630_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_630_end_0 = const()[name = string("op_630_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_630_end_mask_0 = const()[name = string("op_630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_630_squeeze_mask_0 = const()[name = string("op_630_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_630_cast_fp16 = slice_by_index(begin = var_630_begin_0, end = var_630_end_0, end_mask = var_630_end_mask_0, squeeze_mask = var_630_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_630_cast_fp16")]; + bool var_632_transpose_x_1 = const()[name = string("op_632_transpose_x_1"), val = bool(true)]; + bool var_632_transpose_y_1 = const()[name = string("op_632_transpose_y_1"), val = bool(false)]; + tensor var_632_cast_fp16 = matmul(transpose_x = var_632_transpose_x_1, transpose_y = var_632_transpose_y_1, x = var_630_cast_fp16, y = v_new_21_cast_fp16)[name = string("op_632_cast_fp16")]; + tensor state_23_cast_fp16 = add(x = var_628_cast_fp16, y = var_632_cast_fp16)[name = string("state_23_cast_fp16")]; + tensor var_635_begin_0 = const()[name = string("op_635_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_635_end_0 = const()[name = string("op_635_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_635_end_mask_0 = const()[name = string("op_635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_635_squeeze_mask_0 = const()[name = string("op_635_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_635_cast_fp16 = slice_by_index(begin = var_635_begin_0, end = var_635_end_0, end_mask = var_635_end_mask_0, squeeze_mask = var_635_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_635_cast_fp16")]; + tensor var_637_begin_0 = const()[name = string("op_637_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_637_end_0 = const()[name = string("op_637_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_637_end_mask_0 = const()[name = string("op_637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_637_squeeze_mask_0 = const()[name = string("op_637_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_637_cast_fp16 = slice_by_index(begin = var_637_begin_0, end = var_637_end_0, end_mask = var_637_end_mask_0, squeeze_mask = var_637_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_637_cast_fp16")]; + bool var_638_transpose_x_0 = const()[name = string("op_638_transpose_x_0"), val = bool(false)]; + bool var_638_transpose_y_0 = const()[name = string("op_638_transpose_y_0"), val = bool(false)]; + tensor var_638_cast_fp16 = matmul(transpose_x = var_638_transpose_x_0, transpose_y = var_638_transpose_y_0, x = var_637_cast_fp16, y = state_23_cast_fp16)[name = string("op_638_cast_fp16")]; + tensor v_new_23_cast_fp16 = sub(x = var_635_cast_fp16, y = var_638_cast_fp16)[name = string("v_new_23_cast_fp16")]; + tensor var_641_begin_0 = const()[name = string("op_641_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_641_end_0 = const()[name = string("op_641_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_641_end_mask_0 = const()[name = string("op_641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_641_squeeze_mask_0 = const()[name = string("op_641_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_641_cast_fp16 = slice_by_index(begin = var_641_begin_0, end = var_641_end_0, end_mask = var_641_end_mask_0, squeeze_mask = var_641_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_641_cast_fp16")]; + bool var_642_transpose_x_0 = const()[name = string("op_642_transpose_x_0"), val = bool(false)]; + bool var_642_transpose_y_0 = const()[name = string("op_642_transpose_y_0"), val = bool(false)]; + tensor var_642_cast_fp16 = matmul(transpose_x = var_642_transpose_x_0, transpose_y = var_642_transpose_y_0, x = var_641_cast_fp16, y = state_23_cast_fp16)[name = string("op_642_cast_fp16")]; + tensor var_644_begin_0 = const()[name = string("op_644_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_644_end_0 = const()[name = string("op_644_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_644_end_mask_0 = const()[name = string("op_644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_644_squeeze_mask_0 = const()[name = string("op_644_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_644_cast_fp16 = slice_by_index(begin = var_644_begin_0, end = var_644_end_0, end_mask = var_644_end_mask_0, squeeze_mask = var_644_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_644_cast_fp16")]; + bool var_645_transpose_x_0 = const()[name = string("op_645_transpose_x_0"), val = bool(false)]; + bool var_645_transpose_y_0 = const()[name = string("op_645_transpose_y_0"), val = bool(false)]; + tensor var_645_cast_fp16 = matmul(transpose_x = var_645_transpose_x_0, transpose_y = var_645_transpose_y_0, x = var_644_cast_fp16, y = v_new_23_cast_fp16)[name = string("op_645_cast_fp16")]; + tensor var_646_cast_fp16 = add(x = var_642_cast_fp16, y = var_645_cast_fp16)[name = string("op_646_cast_fp16")]; + tensor var_648_begin_0 = const()[name = string("op_648_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_648_end_0 = const()[name = string("op_648_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_648_end_mask_0 = const()[name = string("op_648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_648_squeeze_mask_0 = const()[name = string("op_648_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_648_cast_fp16 = slice_by_index(begin = var_648_begin_0, end = var_648_end_0, end_mask = var_648_end_mask_0, squeeze_mask = var_648_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_648_cast_fp16")]; + tensor var_649_cast_fp16 = mul(x = state_23_cast_fp16, y = var_648_cast_fp16)[name = string("op_649_cast_fp16")]; + tensor var_651_begin_0 = const()[name = string("op_651_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_651_end_0 = const()[name = string("op_651_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_651_end_mask_0 = const()[name = string("op_651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_651_squeeze_mask_0 = const()[name = string("op_651_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_651_cast_fp16 = slice_by_index(begin = var_651_begin_0, end = var_651_end_0, end_mask = var_651_end_mask_0, squeeze_mask = var_651_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_651_cast_fp16")]; + bool var_653_transpose_x_1 = const()[name = string("op_653_transpose_x_1"), val = bool(true)]; + bool var_653_transpose_y_1 = const()[name = string("op_653_transpose_y_1"), val = bool(false)]; + tensor var_653_cast_fp16 = matmul(transpose_x = var_653_transpose_x_1, transpose_y = var_653_transpose_y_1, x = var_651_cast_fp16, y = v_new_23_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor state_25_cast_fp16 = add(x = var_649_cast_fp16, y = var_653_cast_fp16)[name = string("state_25_cast_fp16")]; + tensor var_656_begin_0 = const()[name = string("op_656_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_656_end_0 = const()[name = string("op_656_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_656_end_mask_0 = const()[name = string("op_656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_656_squeeze_mask_0 = const()[name = string("op_656_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_656_cast_fp16 = slice_by_index(begin = var_656_begin_0, end = var_656_end_0, end_mask = var_656_end_mask_0, squeeze_mask = var_656_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_656_cast_fp16")]; + tensor var_658_begin_0 = const()[name = string("op_658_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_658_end_0 = const()[name = string("op_658_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_658_end_mask_0 = const()[name = string("op_658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_658_squeeze_mask_0 = const()[name = string("op_658_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_658_cast_fp16 = slice_by_index(begin = var_658_begin_0, end = var_658_end_0, end_mask = var_658_end_mask_0, squeeze_mask = var_658_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_658_cast_fp16")]; + bool var_659_transpose_x_0 = const()[name = string("op_659_transpose_x_0"), val = bool(false)]; + bool var_659_transpose_y_0 = const()[name = string("op_659_transpose_y_0"), val = bool(false)]; + tensor var_659_cast_fp16 = matmul(transpose_x = var_659_transpose_x_0, transpose_y = var_659_transpose_y_0, x = var_658_cast_fp16, y = state_25_cast_fp16)[name = string("op_659_cast_fp16")]; + tensor v_new_25_cast_fp16 = sub(x = var_656_cast_fp16, y = var_659_cast_fp16)[name = string("v_new_25_cast_fp16")]; + tensor var_662_begin_0 = const()[name = string("op_662_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_662_end_0 = const()[name = string("op_662_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_662_end_mask_0 = const()[name = string("op_662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_662_squeeze_mask_0 = const()[name = string("op_662_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_662_cast_fp16 = slice_by_index(begin = var_662_begin_0, end = var_662_end_0, end_mask = var_662_end_mask_0, squeeze_mask = var_662_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_662_cast_fp16")]; + bool var_663_transpose_x_0 = const()[name = string("op_663_transpose_x_0"), val = bool(false)]; + bool var_663_transpose_y_0 = const()[name = string("op_663_transpose_y_0"), val = bool(false)]; + tensor var_663_cast_fp16 = matmul(transpose_x = var_663_transpose_x_0, transpose_y = var_663_transpose_y_0, x = var_662_cast_fp16, y = state_25_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_665_begin_0 = const()[name = string("op_665_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_665_end_0 = const()[name = string("op_665_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_665_end_mask_0 = const()[name = string("op_665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_665_squeeze_mask_0 = const()[name = string("op_665_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_665_cast_fp16 = slice_by_index(begin = var_665_begin_0, end = var_665_end_0, end_mask = var_665_end_mask_0, squeeze_mask = var_665_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_665_cast_fp16")]; + bool var_666_transpose_x_0 = const()[name = string("op_666_transpose_x_0"), val = bool(false)]; + bool var_666_transpose_y_0 = const()[name = string("op_666_transpose_y_0"), val = bool(false)]; + tensor var_666_cast_fp16 = matmul(transpose_x = var_666_transpose_x_0, transpose_y = var_666_transpose_y_0, x = var_665_cast_fp16, y = v_new_25_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_667_cast_fp16 = add(x = var_663_cast_fp16, y = var_666_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_669_begin_0 = const()[name = string("op_669_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_669_end_0 = const()[name = string("op_669_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_669_end_mask_0 = const()[name = string("op_669_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_669_squeeze_mask_0 = const()[name = string("op_669_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_669_cast_fp16 = slice_by_index(begin = var_669_begin_0, end = var_669_end_0, end_mask = var_669_end_mask_0, squeeze_mask = var_669_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_669_cast_fp16")]; + tensor var_670_cast_fp16 = mul(x = state_25_cast_fp16, y = var_669_cast_fp16)[name = string("op_670_cast_fp16")]; + tensor var_672_begin_0 = const()[name = string("op_672_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_672_end_0 = const()[name = string("op_672_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_672_end_mask_0 = const()[name = string("op_672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_672_squeeze_mask_0 = const()[name = string("op_672_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_672_cast_fp16 = slice_by_index(begin = var_672_begin_0, end = var_672_end_0, end_mask = var_672_end_mask_0, squeeze_mask = var_672_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_672_cast_fp16")]; + bool var_674_transpose_x_1 = const()[name = string("op_674_transpose_x_1"), val = bool(true)]; + bool var_674_transpose_y_1 = const()[name = string("op_674_transpose_y_1"), val = bool(false)]; + tensor var_674_cast_fp16 = matmul(transpose_x = var_674_transpose_x_1, transpose_y = var_674_transpose_y_1, x = var_672_cast_fp16, y = v_new_25_cast_fp16)[name = string("op_674_cast_fp16")]; + tensor state_27_cast_fp16 = add(x = var_670_cast_fp16, y = var_674_cast_fp16)[name = string("state_27_cast_fp16")]; + tensor var_677_begin_0 = const()[name = string("op_677_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_677_end_0 = const()[name = string("op_677_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_677_end_mask_0 = const()[name = string("op_677_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_677_squeeze_mask_0 = const()[name = string("op_677_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_677_cast_fp16 = slice_by_index(begin = var_677_begin_0, end = var_677_end_0, end_mask = var_677_end_mask_0, squeeze_mask = var_677_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_677_cast_fp16")]; + tensor var_679_begin_0 = const()[name = string("op_679_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_679_end_0 = const()[name = string("op_679_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_679_end_mask_0 = const()[name = string("op_679_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_679_squeeze_mask_0 = const()[name = string("op_679_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_679_cast_fp16 = slice_by_index(begin = var_679_begin_0, end = var_679_end_0, end_mask = var_679_end_mask_0, squeeze_mask = var_679_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_679_cast_fp16")]; + bool var_680_transpose_x_0 = const()[name = string("op_680_transpose_x_0"), val = bool(false)]; + bool var_680_transpose_y_0 = const()[name = string("op_680_transpose_y_0"), val = bool(false)]; + tensor var_680_cast_fp16 = matmul(transpose_x = var_680_transpose_x_0, transpose_y = var_680_transpose_y_0, x = var_679_cast_fp16, y = state_27_cast_fp16)[name = string("op_680_cast_fp16")]; + tensor v_new_27_cast_fp16 = sub(x = var_677_cast_fp16, y = var_680_cast_fp16)[name = string("v_new_27_cast_fp16")]; + tensor var_683_begin_0 = const()[name = string("op_683_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_683_end_0 = const()[name = string("op_683_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_683_end_mask_0 = const()[name = string("op_683_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_683_squeeze_mask_0 = const()[name = string("op_683_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_683_cast_fp16 = slice_by_index(begin = var_683_begin_0, end = var_683_end_0, end_mask = var_683_end_mask_0, squeeze_mask = var_683_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_683_cast_fp16")]; + bool var_684_transpose_x_0 = const()[name = string("op_684_transpose_x_0"), val = bool(false)]; + bool var_684_transpose_y_0 = const()[name = string("op_684_transpose_y_0"), val = bool(false)]; + tensor var_684_cast_fp16 = matmul(transpose_x = var_684_transpose_x_0, transpose_y = var_684_transpose_y_0, x = var_683_cast_fp16, y = state_27_cast_fp16)[name = string("op_684_cast_fp16")]; + tensor var_686_begin_0 = const()[name = string("op_686_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_686_end_0 = const()[name = string("op_686_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_686_end_mask_0 = const()[name = string("op_686_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_686_squeeze_mask_0 = const()[name = string("op_686_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_686_cast_fp16 = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, squeeze_mask = var_686_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_686_cast_fp16")]; + bool var_687_transpose_x_0 = const()[name = string("op_687_transpose_x_0"), val = bool(false)]; + bool var_687_transpose_y_0 = const()[name = string("op_687_transpose_y_0"), val = bool(false)]; + tensor var_687_cast_fp16 = matmul(transpose_x = var_687_transpose_x_0, transpose_y = var_687_transpose_y_0, x = var_686_cast_fp16, y = v_new_27_cast_fp16)[name = string("op_687_cast_fp16")]; + tensor var_688_cast_fp16 = add(x = var_684_cast_fp16, y = var_687_cast_fp16)[name = string("op_688_cast_fp16")]; + tensor var_690_begin_0 = const()[name = string("op_690_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_690_end_0 = const()[name = string("op_690_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_690_end_mask_0 = const()[name = string("op_690_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_690_squeeze_mask_0 = const()[name = string("op_690_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_690_cast_fp16 = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, squeeze_mask = var_690_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_690_cast_fp16")]; + tensor var_691_cast_fp16 = mul(x = state_27_cast_fp16, y = var_690_cast_fp16)[name = string("op_691_cast_fp16")]; + tensor var_693_begin_0 = const()[name = string("op_693_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_693_end_0 = const()[name = string("op_693_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_693_end_mask_0 = const()[name = string("op_693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_693_squeeze_mask_0 = const()[name = string("op_693_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_693_cast_fp16 = slice_by_index(begin = var_693_begin_0, end = var_693_end_0, end_mask = var_693_end_mask_0, squeeze_mask = var_693_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_693_cast_fp16")]; + bool var_695_transpose_x_1 = const()[name = string("op_695_transpose_x_1"), val = bool(true)]; + bool var_695_transpose_y_1 = const()[name = string("op_695_transpose_y_1"), val = bool(false)]; + tensor var_695_cast_fp16 = matmul(transpose_x = var_695_transpose_x_1, transpose_y = var_695_transpose_y_1, x = var_693_cast_fp16, y = v_new_27_cast_fp16)[name = string("op_695_cast_fp16")]; + tensor state_29_cast_fp16 = add(x = var_691_cast_fp16, y = var_695_cast_fp16)[name = string("state_29_cast_fp16")]; + tensor var_698_begin_0 = const()[name = string("op_698_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_698_end_0 = const()[name = string("op_698_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_698_end_mask_0 = const()[name = string("op_698_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_698_squeeze_mask_0 = const()[name = string("op_698_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_698_cast_fp16 = slice_by_index(begin = var_698_begin_0, end = var_698_end_0, end_mask = var_698_end_mask_0, squeeze_mask = var_698_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_698_cast_fp16")]; + tensor var_700_begin_0 = const()[name = string("op_700_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_700_end_0 = const()[name = string("op_700_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_700_end_mask_0 = const()[name = string("op_700_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_700_squeeze_mask_0 = const()[name = string("op_700_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_700_cast_fp16 = slice_by_index(begin = var_700_begin_0, end = var_700_end_0, end_mask = var_700_end_mask_0, squeeze_mask = var_700_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_700_cast_fp16")]; + bool var_701_transpose_x_0 = const()[name = string("op_701_transpose_x_0"), val = bool(false)]; + bool var_701_transpose_y_0 = const()[name = string("op_701_transpose_y_0"), val = bool(false)]; + tensor var_701_cast_fp16 = matmul(transpose_x = var_701_transpose_x_0, transpose_y = var_701_transpose_y_0, x = var_700_cast_fp16, y = state_29_cast_fp16)[name = string("op_701_cast_fp16")]; + tensor v_new_29_cast_fp16 = sub(x = var_698_cast_fp16, y = var_701_cast_fp16)[name = string("v_new_29_cast_fp16")]; + tensor var_704_begin_0 = const()[name = string("op_704_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_704_end_0 = const()[name = string("op_704_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_704_end_mask_0 = const()[name = string("op_704_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_704_squeeze_mask_0 = const()[name = string("op_704_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_704_cast_fp16 = slice_by_index(begin = var_704_begin_0, end = var_704_end_0, end_mask = var_704_end_mask_0, squeeze_mask = var_704_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_704_cast_fp16")]; + bool var_705_transpose_x_0 = const()[name = string("op_705_transpose_x_0"), val = bool(false)]; + bool var_705_transpose_y_0 = const()[name = string("op_705_transpose_y_0"), val = bool(false)]; + tensor var_705_cast_fp16 = matmul(transpose_x = var_705_transpose_x_0, transpose_y = var_705_transpose_y_0, x = var_704_cast_fp16, y = state_29_cast_fp16)[name = string("op_705_cast_fp16")]; + tensor var_707_begin_0 = const()[name = string("op_707_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_707_end_0 = const()[name = string("op_707_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_707_end_mask_0 = const()[name = string("op_707_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_707_squeeze_mask_0 = const()[name = string("op_707_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, squeeze_mask = var_707_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_707_cast_fp16")]; + bool var_708_transpose_x_0 = const()[name = string("op_708_transpose_x_0"), val = bool(false)]; + bool var_708_transpose_y_0 = const()[name = string("op_708_transpose_y_0"), val = bool(false)]; + tensor var_708_cast_fp16 = matmul(transpose_x = var_708_transpose_x_0, transpose_y = var_708_transpose_y_0, x = var_707_cast_fp16, y = v_new_29_cast_fp16)[name = string("op_708_cast_fp16")]; + tensor var_709_cast_fp16 = add(x = var_705_cast_fp16, y = var_708_cast_fp16)[name = string("op_709_cast_fp16")]; + tensor var_711_begin_0 = const()[name = string("op_711_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_711_end_0 = const()[name = string("op_711_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_711_end_mask_0 = const()[name = string("op_711_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_711_squeeze_mask_0 = const()[name = string("op_711_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_711_cast_fp16 = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, squeeze_mask = var_711_squeeze_mask_0, x = chunk_decay_1_cast_fp16)[name = string("op_711_cast_fp16")]; + tensor var_712_cast_fp16 = mul(x = state_29_cast_fp16, y = var_711_cast_fp16)[name = string("op_712_cast_fp16")]; + tensor var_714_begin_0 = const()[name = string("op_714_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_714_end_0 = const()[name = string("op_714_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_714_end_mask_0 = const()[name = string("op_714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_714_squeeze_mask_0 = const()[name = string("op_714_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_714_cast_fp16 = slice_by_index(begin = var_714_begin_0, end = var_714_end_0, end_mask = var_714_end_mask_0, squeeze_mask = var_714_squeeze_mask_0, x = k_dec_1_cast_fp16)[name = string("op_714_cast_fp16")]; + bool var_716_transpose_x_1 = const()[name = string("op_716_transpose_x_1"), val = bool(true)]; + bool var_716_transpose_y_1 = const()[name = string("op_716_transpose_y_1"), val = bool(false)]; + tensor var_716_cast_fp16 = matmul(transpose_x = var_716_transpose_x_1, transpose_y = var_716_transpose_y_1, x = var_714_cast_fp16, y = v_new_29_cast_fp16)[name = string("op_716_cast_fp16")]; + tensor state_31_cast_fp16 = add(x = var_712_cast_fp16, y = var_716_cast_fp16)[name = string("state_31_cast_fp16")]; + tensor var_719_begin_0 = const()[name = string("op_719_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_719_end_0 = const()[name = string("op_719_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_719_end_mask_0 = const()[name = string("op_719_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_719_squeeze_mask_0 = const()[name = string("op_719_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_719_cast_fp16 = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, squeeze_mask = var_719_squeeze_mask_0, x = new_v_1_cast_fp16)[name = string("op_719_cast_fp16")]; + tensor var_721_begin_0 = const()[name = string("op_721_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_721_end_0 = const()[name = string("op_721_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_721_end_mask_0 = const()[name = string("op_721_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_721_squeeze_mask_0 = const()[name = string("op_721_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_721_cast_fp16 = slice_by_index(begin = var_721_begin_0, end = var_721_end_0, end_mask = var_721_end_mask_0, squeeze_mask = var_721_squeeze_mask_0, x = k_cumdecay_1_cast_fp16)[name = string("op_721_cast_fp16")]; + bool var_722_transpose_x_0 = const()[name = string("op_722_transpose_x_0"), val = bool(false)]; + bool var_722_transpose_y_0 = const()[name = string("op_722_transpose_y_0"), val = bool(false)]; + tensor var_722_cast_fp16 = matmul(transpose_x = var_722_transpose_x_0, transpose_y = var_722_transpose_y_0, x = var_721_cast_fp16, y = state_31_cast_fp16)[name = string("op_722_cast_fp16")]; + tensor v_new_31_cast_fp16 = sub(x = var_719_cast_fp16, y = var_722_cast_fp16)[name = string("v_new_31_cast_fp16")]; + tensor var_725_begin_0 = const()[name = string("op_725_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_725_end_0 = const()[name = string("op_725_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_725_end_mask_0 = const()[name = string("op_725_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_725_squeeze_mask_0 = const()[name = string("op_725_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, squeeze_mask = var_725_squeeze_mask_0, x = q_dec_1_cast_fp16)[name = string("op_725_cast_fp16")]; + bool var_726_transpose_x_0 = const()[name = string("op_726_transpose_x_0"), val = bool(false)]; + bool var_726_transpose_y_0 = const()[name = string("op_726_transpose_y_0"), val = bool(false)]; + tensor var_726_cast_fp16 = matmul(transpose_x = var_726_transpose_x_0, transpose_y = var_726_transpose_y_0, x = var_725_cast_fp16, y = state_31_cast_fp16)[name = string("op_726_cast_fp16")]; + tensor var_728_begin_0 = const()[name = string("op_728_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_728_end_0 = const()[name = string("op_728_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_728_end_mask_0 = const()[name = string("op_728_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_728_squeeze_mask_0 = const()[name = string("op_728_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_728_cast_fp16 = slice_by_index(begin = var_728_begin_0, end = var_728_end_0, end_mask = var_728_end_mask_0, squeeze_mask = var_728_squeeze_mask_0, x = attn_intra_1_cast_fp16)[name = string("op_728_cast_fp16")]; + bool var_729_transpose_x_0 = const()[name = string("op_729_transpose_x_0"), val = bool(false)]; + bool var_729_transpose_y_0 = const()[name = string("op_729_transpose_y_0"), val = bool(false)]; + tensor var_729_cast_fp16 = matmul(transpose_x = var_729_transpose_x_0, transpose_y = var_729_transpose_y_0, x = var_728_cast_fp16, y = v_new_31_cast_fp16)[name = string("op_729_cast_fp16")]; + tensor var_730_cast_fp16 = add(x = var_726_cast_fp16, y = var_729_cast_fp16)[name = string("op_730_cast_fp16")]; + int32 var_732_axis_0 = const()[name = string("op_732_axis_0"), val = int32(1)]; + tensor var_732_cast_fp16 = stack(axis = var_732_axis_0, values = (var_415_cast_fp16, var_436_cast_fp16, var_457_cast_fp16, var_478_cast_fp16, var_499_cast_fp16, var_520_cast_fp16, var_541_cast_fp16, var_562_cast_fp16, var_583_cast_fp16, var_604_cast_fp16, var_625_cast_fp16, var_646_cast_fp16, var_667_cast_fp16, var_688_cast_fp16, var_709_cast_fp16, var_730_cast_fp16))[name = string("op_732_cast_fp16")]; + tensor var_733 = const()[name = string("op_733"), val = tensor([32, 1024, 128])]; + tensor var_734_cast_fp16 = reshape(shape = var_733, x = var_732_cast_fp16)[name = string("op_734_cast_fp16")]; + tensor var_735 = const()[name = string("op_735"), val = tensor([1, 0, 2])]; + tensor input_7_begin_0 = const()[name = string("input_7_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_end_0 = const()[name = string("input_7_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_7_end_mask_0 = const()[name = string("input_7_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_7_squeeze_mask_0 = const()[name = string("input_7_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_7_cast_fp16 = slice_by_index(begin = input_7_begin_0, end = input_7_end_0, end_mask = input_7_end_mask_0, squeeze_mask = input_7_squeeze_mask_0, x = z_1_cast_fp16)[name = string("input_7_cast_fp16")]; + fp16 var_57_promoted_1_to_fp16 = const()[name = string("op_57_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_17_cast_fp16 = transpose(perm = var_735, x = var_734_cast_fp16)[name = string("transpose_48")]; + tensor var_739_cast_fp16 = pow(x = x_17_cast_fp16, y = var_57_promoted_1_to_fp16)[name = string("op_739_cast_fp16")]; + tensor var_741_axes_0 = const()[name = string("op_741_axes_0"), val = tensor([-1])]; + bool var_741_keep_dims_0 = const()[name = string("op_741_keep_dims_0"), val = bool(true)]; + tensor var_741_cast_fp16 = reduce_mean(axes = var_741_axes_0, keep_dims = var_741_keep_dims_0, x = var_739_cast_fp16)[name = string("op_741_cast_fp16")]; + fp16 var_742_to_fp16 = const()[name = string("op_742_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_743_cast_fp16 = add(x = var_741_cast_fp16, y = var_742_to_fp16)[name = string("op_743_cast_fp16")]; + fp32 var_744_epsilon_0 = const()[name = string("op_744_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_744_cast_fp16 = rsqrt(epsilon = var_744_epsilon_0, x = var_743_cast_fp16)[name = string("op_744_cast_fp16")]; + tensor x_19_cast_fp16 = mul(x = x_17_cast_fp16, y = var_744_cast_fp16)[name = string("x_19_cast_fp16")]; + tensor layers_0_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_0_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67077824)))]; + tensor var_746_cast_fp16 = mul(x = layers_0_linear_attn_norm_weight_to_fp16, y = x_19_cast_fp16)[name = string("op_746_cast_fp16")]; + tensor var_747_cast_fp16 = silu(x = input_7_cast_fp16)[name = string("op_747_cast_fp16")]; + tensor core_1_cast_fp16 = mul(x = var_746_cast_fp16, y = var_747_cast_fp16)[name = string("core_1_cast_fp16")]; + tensor var_749 = const()[name = string("op_749"), val = tensor([1, 1024, 4096])]; + tensor input_9_cast_fp16 = reshape(shape = var_749, x = core_1_cast_fp16)[name = string("input_9_cast_fp16")]; + tensor layers_0_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_0_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67078144)))]; + tensor linear_4_bias_0_to_fp16 = const()[name = string("linear_4_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88049728)))]; + tensor linear_4_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_0_linear_attn_out_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_4_cast_fp16")]; + tensor x_21_cast_fp16 = add(x = hidden, y = linear_4_cast_fp16)[name = string("x_21_cast_fp16")]; + fp16 var_57_promoted_2_to_fp16 = const()[name = string("op_57_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_755_cast_fp16 = pow(x = x_21_cast_fp16, y = var_57_promoted_2_to_fp16)[name = string("op_755_cast_fp16")]; + tensor var_757_axes_0 = const()[name = string("op_757_axes_0"), val = tensor([-1])]; + bool var_757_keep_dims_0 = const()[name = string("op_757_keep_dims_0"), val = bool(true)]; + tensor var_757_cast_fp16 = reduce_mean(axes = var_757_axes_0, keep_dims = var_757_keep_dims_0, x = var_755_cast_fp16)[name = string("op_757_cast_fp16")]; + fp16 var_758_to_fp16 = const()[name = string("op_758_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_759_cast_fp16 = add(x = var_757_cast_fp16, y = var_758_to_fp16)[name = string("op_759_cast_fp16")]; + fp32 var_760_epsilon_0 = const()[name = string("op_760_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_760_cast_fp16 = rsqrt(epsilon = var_760_epsilon_0, x = var_759_cast_fp16)[name = string("op_760_cast_fp16")]; + tensor x_23_cast_fp16 = mul(x = x_21_cast_fp16, y = var_760_cast_fp16)[name = string("x_23_cast_fp16")]; + tensor var_763_to_fp16 = const()[name = string("op_763_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88054912)))]; + tensor input_11_cast_fp16 = mul(x = x_23_cast_fp16, y = var_763_to_fp16)[name = string("input_11_cast_fp16")]; + tensor layers_0_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88060096)))]; + tensor linear_5_bias_0_to_fp16 = const()[name = string("linear_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135246080)))]; + tensor linear_5_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_0_mlp_gate_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_5_cast_fp16")]; + tensor var_770_cast_fp16 = silu(x = linear_5_cast_fp16)[name = string("op_770_cast_fp16")]; + tensor layers_0_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135264576)))]; + tensor linear_6_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_0_mlp_up_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_6_cast_fp16")]; + tensor input_15_cast_fp16 = mul(x = var_770_cast_fp16, y = linear_6_cast_fp16)[name = string("input_15_cast_fp16")]; + tensor layers_0_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182450560)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_0_mlp_down_proj_weight_to_fp16, x = input_15_cast_fp16)[name = string("linear_7_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_21_cast_fp16, y = linear_7_cast_fp16)[name = string("x_25_cast_fp16")]; + int32 var_790 = const()[name = string("op_790"), val = int32(-2)]; + int32 var_811 = const()[name = string("op_811"), val = int32(-1)]; + fp16 var_810_promoted_to_fp16 = const()[name = string("op_810_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_820_cast_fp16 = pow(x = x_25_cast_fp16, y = var_810_promoted_to_fp16)[name = string("op_820_cast_fp16")]; + tensor var_822_axes_0 = const()[name = string("op_822_axes_0"), val = tensor([-1])]; + bool var_822_keep_dims_0 = const()[name = string("op_822_keep_dims_0"), val = bool(true)]; + tensor var_822_cast_fp16 = reduce_mean(axes = var_822_axes_0, keep_dims = var_822_keep_dims_0, x = var_820_cast_fp16)[name = string("op_822_cast_fp16")]; + fp16 var_823_to_fp16 = const()[name = string("op_823_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_824_cast_fp16 = add(x = var_822_cast_fp16, y = var_823_to_fp16)[name = string("op_824_cast_fp16")]; + fp32 var_825_epsilon_0 = const()[name = string("op_825_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_825_cast_fp16 = rsqrt(epsilon = var_825_epsilon_0, x = var_824_cast_fp16)[name = string("op_825_cast_fp16")]; + tensor x_27_cast_fp16 = mul(x = x_25_cast_fp16, y = var_825_cast_fp16)[name = string("x_27_cast_fp16")]; + tensor var_828_to_fp16 = const()[name = string("op_828_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229636544)))]; + tensor x_29_cast_fp16 = mul(x = x_27_cast_fp16, y = var_828_to_fp16)[name = string("x_29_cast_fp16")]; + tensor layers_1_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_1_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229641728)))]; + tensor linear_8_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_linear_attn_in_proj_qkv_weight_to_fp16, x = x_29_cast_fp16)[name = string("linear_8_cast_fp16")]; + tensor input_17_perm_0 = const()[name = string("input_17_perm_0"), val = tensor([0, 2, 1])]; + tensor input_19_pad_0 = const()[name = string("input_19_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_19_mode_0 = const()[name = string("input_19_mode_0"), val = string("constant")]; + fp16 const_13_to_fp16 = const()[name = string("const_13_to_fp16"), val = fp16(0x0p+0)]; + tensor input_17_cast_fp16 = transpose(perm = input_17_perm_0, x = linear_8_cast_fp16)[name = string("transpose_47")]; + tensor input_19_cast_fp16 = pad(constant_val = const_13_to_fp16, mode = input_19_mode_0, pad = input_19_pad_0, x = input_17_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(8192)]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1])]; + tensor layers_1_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_1_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271584832)))]; + tensor input_21_cast_fp16 = conv(dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_1_linear_attn_conv1d_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + tensor var_857_cast_fp16 = silu(x = input_21_cast_fp16)[name = string("op_857_cast_fp16")]; + tensor qkv_3_perm_0 = const()[name = string("qkv_3_perm_0"), val = tensor([0, 2, 1])]; + tensor var_859_begin_0 = const()[name = string("op_859_begin_0"), val = tensor([0, 0, 0])]; + tensor var_859_end_0 = const()[name = string("op_859_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_859_end_mask_0 = const()[name = string("op_859_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_3_cast_fp16 = transpose(perm = qkv_3_perm_0, x = var_857_cast_fp16)[name = string("transpose_46")]; + tensor var_859_cast_fp16 = slice_by_index(begin = var_859_begin_0, end = var_859_end_0, end_mask = var_859_end_mask_0, x = qkv_3_cast_fp16)[name = string("op_859_cast_fp16")]; + tensor var_860 = const()[name = string("op_860"), val = tensor([1, 1024, 16, 128])]; + tensor q_11_cast_fp16 = reshape(shape = var_860, x = var_859_cast_fp16)[name = string("q_11_cast_fp16")]; + tensor var_862_begin_0 = const()[name = string("op_862_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_862_end_0 = const()[name = string("op_862_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_862_end_mask_0 = const()[name = string("op_862_end_mask_0"), val = tensor([true, true, false])]; + tensor var_862_cast_fp16 = slice_by_index(begin = var_862_begin_0, end = var_862_end_0, end_mask = var_862_end_mask_0, x = qkv_3_cast_fp16)[name = string("op_862_cast_fp16")]; + tensor var_863 = const()[name = string("op_863"), val = tensor([1, 1024, 16, 128])]; + tensor k_9_cast_fp16 = reshape(shape = var_863, x = var_862_cast_fp16)[name = string("k_9_cast_fp16")]; + tensor var_865_begin_0 = const()[name = string("op_865_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_865_end_0 = const()[name = string("op_865_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_865_end_mask_0 = const()[name = string("op_865_end_mask_0"), val = tensor([true, true, true])]; + tensor var_865_cast_fp16 = slice_by_index(begin = var_865_begin_0, end = var_865_end_0, end_mask = var_865_end_mask_0, x = qkv_3_cast_fp16)[name = string("op_865_cast_fp16")]; + tensor var_866 = const()[name = string("op_866"), val = tensor([1, 1024, 32, 128])]; + tensor v_5_cast_fp16 = reshape(shape = var_866, x = var_865_cast_fp16)[name = string("v_5_cast_fp16")]; + tensor layers_1_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_1_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271650432)))]; + tensor linear_9_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_1_linear_attn_in_proj_z_weight_to_fp16, x = x_29_cast_fp16)[name = string("linear_9_cast_fp16")]; + tensor var_870 = const()[name = string("op_870"), val = tensor([1, 1024, 32, 128])]; + tensor z_3_cast_fp16 = reshape(shape = var_870, x = linear_9_cast_fp16)[name = string("z_3_cast_fp16")]; + tensor layers_1_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_1_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292622016)))]; + tensor linear_10_bias_0_to_fp16 = const()[name = string("linear_10_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292785920)))]; + tensor linear_10_cast_fp16 = linear(bias = linear_10_bias_0_to_fp16, weight = layers_1_linear_attn_in_proj_b_weight_to_fp16, x = x_29_cast_fp16)[name = string("linear_10_cast_fp16")]; + tensor beta_5_cast_fp16 = sigmoid(x = linear_10_cast_fp16)[name = string("beta_5_cast_fp16")]; + tensor var_879_weight_0_to_fp16 = const()[name = string("op_879_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292786048)))]; + tensor var_879_bias_0_to_fp16 = const()[name = string("op_879_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292949952)))]; + tensor var_879_cast_fp16 = linear(bias = var_879_bias_0_to_fp16, weight = var_879_weight_0_to_fp16, x = x_29_cast_fp16)[name = string("op_879_cast_fp16")]; + tensor var_880_cast_fp16 = softplus(x = var_879_cast_fp16)[name = string("op_880_cast_fp16")]; + tensor var_876_to_fp16 = const()[name = string("op_876_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292950080)))]; + tensor g_5_cast_fp16 = mul(x = var_876_to_fp16, y = var_880_cast_fp16)[name = string("g_5_cast_fp16")]; + tensor var_882_cast_fp16 = mul(x = q_11_cast_fp16, y = q_11_cast_fp16)[name = string("op_882_cast_fp16")]; + tensor var_884_axes_0 = const()[name = string("op_884_axes_0"), val = tensor([-1])]; + bool var_884_keep_dims_0 = const()[name = string("op_884_keep_dims_0"), val = bool(true)]; + tensor var_884_cast_fp16 = reduce_sum(axes = var_884_axes_0, keep_dims = var_884_keep_dims_0, x = var_882_cast_fp16)[name = string("op_884_cast_fp16")]; + fp16 var_885_to_fp16 = const()[name = string("op_885_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_886_cast_fp16 = add(x = var_884_cast_fp16, y = var_885_to_fp16)[name = string("op_886_cast_fp16")]; + fp32 var_887_epsilon_0 = const()[name = string("op_887_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_887_cast_fp16 = rsqrt(epsilon = var_887_epsilon_0, x = var_886_cast_fp16)[name = string("op_887_cast_fp16")]; + tensor q_13_cast_fp16 = mul(x = q_11_cast_fp16, y = var_887_cast_fp16)[name = string("q_13_cast_fp16")]; + tensor var_889_cast_fp16 = mul(x = k_9_cast_fp16, y = k_9_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_891_axes_0 = const()[name = string("op_891_axes_0"), val = tensor([-1])]; + bool var_891_keep_dims_0 = const()[name = string("op_891_keep_dims_0"), val = bool(true)]; + tensor var_891_cast_fp16 = reduce_sum(axes = var_891_axes_0, keep_dims = var_891_keep_dims_0, x = var_889_cast_fp16)[name = string("op_891_cast_fp16")]; + fp16 var_892_to_fp16 = const()[name = string("op_892_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_893_cast_fp16 = add(x = var_891_cast_fp16, y = var_892_to_fp16)[name = string("op_893_cast_fp16")]; + fp32 var_894_epsilon_0 = const()[name = string("op_894_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_894_cast_fp16 = rsqrt(epsilon = var_894_epsilon_0, x = var_893_cast_fp16)[name = string("op_894_cast_fp16")]; + tensor k_11_cast_fp16 = mul(x = k_9_cast_fp16, y = var_894_cast_fp16)[name = string("k_11_cast_fp16")]; + tensor var_899_axes_0 = const()[name = string("op_899_axes_0"), val = tensor([3])]; + tensor var_899_cast_fp16 = expand_dims(axes = var_899_axes_0, x = q_13_cast_fp16)[name = string("op_899_cast_fp16")]; + tensor var_901_reps_0 = const()[name = string("op_901_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_901_cast_fp16 = tile(reps = var_901_reps_0, x = var_899_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_902 = const()[name = string("op_902"), val = tensor([1, 1024, 32, 128])]; + tensor q_15_cast_fp16 = reshape(shape = var_902, x = var_901_cast_fp16)[name = string("q_15_cast_fp16")]; + tensor var_907_axes_0 = const()[name = string("op_907_axes_0"), val = tensor([3])]; + tensor var_907_cast_fp16 = expand_dims(axes = var_907_axes_0, x = k_11_cast_fp16)[name = string("op_907_cast_fp16")]; + tensor var_909_reps_0 = const()[name = string("op_909_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_909_cast_fp16 = tile(reps = var_909_reps_0, x = var_907_cast_fp16)[name = string("op_909_cast_fp16")]; + tensor var_910 = const()[name = string("op_910"), val = tensor([1, 1024, 32, 128])]; + tensor k_13_cast_fp16 = reshape(shape = var_910, x = var_909_cast_fp16)[name = string("k_13_cast_fp16")]; + fp16 var_912_to_fp16 = const()[name = string("op_912_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_17_cast_fp16 = mul(x = q_15_cast_fp16, y = var_912_to_fp16)[name = string("q_17_cast_fp16")]; + tensor var_914_begin_0 = const()[name = string("op_914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_914_end_0 = const()[name = string("op_914_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_914_end_mask_0 = const()[name = string("op_914_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_914_squeeze_mask_0 = const()[name = string("op_914_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_914_cast_fp16 = slice_by_index(begin = var_914_begin_0, end = var_914_end_0, end_mask = var_914_end_mask_0, squeeze_mask = var_914_squeeze_mask_0, x = q_17_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_915 = const()[name = string("op_915"), val = tensor([1, 0, 2])]; + tensor var_917 = const()[name = string("op_917"), val = tensor([32, 16, 64, 128])]; + tensor var_916_cast_fp16 = transpose(perm = var_915, x = var_914_cast_fp16)[name = string("transpose_45")]; + tensor q_19_cast_fp16 = reshape(shape = var_917, x = var_916_cast_fp16)[name = string("q_19_cast_fp16")]; + tensor var_919_begin_0 = const()[name = string("op_919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_919_end_0 = const()[name = string("op_919_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_919_end_mask_0 = const()[name = string("op_919_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_919_squeeze_mask_0 = const()[name = string("op_919_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_919_cast_fp16 = slice_by_index(begin = var_919_begin_0, end = var_919_end_0, end_mask = var_919_end_mask_0, squeeze_mask = var_919_squeeze_mask_0, x = k_13_cast_fp16)[name = string("op_919_cast_fp16")]; + tensor var_920 = const()[name = string("op_920"), val = tensor([1, 0, 2])]; + tensor var_922 = const()[name = string("op_922"), val = tensor([32, 16, 64, 128])]; + tensor var_921_cast_fp16 = transpose(perm = var_920, x = var_919_cast_fp16)[name = string("transpose_44")]; + tensor k_15_cast_fp16 = reshape(shape = var_922, x = var_921_cast_fp16)[name = string("k_15_cast_fp16")]; + tensor var_924_begin_0 = const()[name = string("op_924_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_924_end_0 = const()[name = string("op_924_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_924_end_mask_0 = const()[name = string("op_924_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_924_squeeze_mask_0 = const()[name = string("op_924_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_924_cast_fp16 = slice_by_index(begin = var_924_begin_0, end = var_924_end_0, end_mask = var_924_end_mask_0, squeeze_mask = var_924_squeeze_mask_0, x = v_5_cast_fp16)[name = string("op_924_cast_fp16")]; + tensor var_925 = const()[name = string("op_925"), val = tensor([1, 0, 2])]; + tensor var_927 = const()[name = string("op_927"), val = tensor([32, 16, 64, 128])]; + tensor var_926_cast_fp16 = transpose(perm = var_925, x = var_924_cast_fp16)[name = string("transpose_43")]; + tensor v_7_cast_fp16 = reshape(shape = var_927, x = var_926_cast_fp16)[name = string("v_7_cast_fp16")]; + tensor var_929_begin_0 = const()[name = string("op_929_begin_0"), val = tensor([0, 0, 0])]; + tensor var_929_end_0 = const()[name = string("op_929_end_0"), val = tensor([1, 1024, 32])]; + tensor var_929_end_mask_0 = const()[name = string("op_929_end_mask_0"), val = tensor([false, true, true])]; + tensor var_929_squeeze_mask_0 = const()[name = string("op_929_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_929_cast_fp16 = slice_by_index(begin = var_929_begin_0, end = var_929_end_0, end_mask = var_929_end_mask_0, squeeze_mask = var_929_squeeze_mask_0, x = beta_5_cast_fp16)[name = string("op_929_cast_fp16")]; + tensor var_930_perm_0 = const()[name = string("op_930_perm_0"), val = tensor([1, 0])]; + tensor var_931 = const()[name = string("op_931"), val = tensor([32, 16, 64, 1])]; + tensor var_930_cast_fp16 = transpose(perm = var_930_perm_0, x = var_929_cast_fp16)[name = string("transpose_42")]; + tensor beta_7_cast_fp16 = reshape(shape = var_931, x = var_930_cast_fp16)[name = string("beta_7_cast_fp16")]; + tensor var_933_begin_0 = const()[name = string("op_933_begin_0"), val = tensor([0, 0, 0])]; + tensor var_933_end_0 = const()[name = string("op_933_end_0"), val = tensor([1, 1024, 32])]; + tensor var_933_end_mask_0 = const()[name = string("op_933_end_mask_0"), val = tensor([false, true, true])]; + tensor var_933_squeeze_mask_0 = const()[name = string("op_933_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_933_cast_fp16 = slice_by_index(begin = var_933_begin_0, end = var_933_end_0, end_mask = var_933_end_mask_0, squeeze_mask = var_933_squeeze_mask_0, x = g_5_cast_fp16)[name = string("op_933_cast_fp16")]; + tensor var_934_perm_0 = const()[name = string("op_934_perm_0"), val = tensor([1, 0])]; + tensor var_935 = const()[name = string("op_935"), val = tensor([32, 16, 64])]; + tensor var_934_cast_fp16 = transpose(perm = var_934_perm_0, x = var_933_cast_fp16)[name = string("transpose_41")]; + tensor g_7_cast_fp16 = reshape(shape = var_935, x = var_934_cast_fp16)[name = string("g_7_cast_fp16")]; + tensor g_col_3_axes_0 = const()[name = string("g_col_3_axes_0"), val = tensor([-1])]; + tensor g_col_3_cast_fp16 = expand_dims(axes = g_col_3_axes_0, x = g_7_cast_fp16)[name = string("g_col_3_cast_fp16")]; + bool var_938_transpose_x_0 = const()[name = string("op_938_transpose_x_0"), val = bool(false)]; + bool var_938_transpose_y_0 = const()[name = string("op_938_transpose_y_0"), val = bool(false)]; + tensor var_938_cast_fp16 = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_3_cast_fp16)[name = string("op_938_cast_fp16")]; + tensor cum_3_axes_0 = const()[name = string("cum_3_axes_0"), val = tensor([-1])]; + tensor cum_3_cast_fp16 = squeeze(axes = cum_3_axes_0, x = var_938_cast_fp16)[name = string("cum_3_cast_fp16")]; + bool var_940_transpose_x_0 = const()[name = string("op_940_transpose_x_0"), val = bool(false)]; + bool var_940_transpose_y_0 = const()[name = string("op_940_transpose_y_0"), val = bool(false)]; + tensor var_940_cast_fp16 = matmul(transpose_x = var_940_transpose_x_0, transpose_y = var_940_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_3_cast_fp16)[name = string("op_940_cast_fp16")]; + tensor to_end_3_axes_0 = const()[name = string("to_end_3_axes_0"), val = tensor([-1])]; + tensor to_end_3_cast_fp16 = squeeze(axes = to_end_3_axes_0, x = var_940_cast_fp16)[name = string("to_end_3_cast_fp16")]; + bool var_942_transpose_x_0 = const()[name = string("op_942_transpose_x_0"), val = bool(false)]; + bool var_942_transpose_y_0 = const()[name = string("op_942_transpose_y_0"), val = bool(false)]; + tensor var_942_cast_fp16 = matmul(transpose_x = var_942_transpose_x_0, transpose_y = var_942_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_3_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_943 = const()[name = string("op_943"), val = tensor([32, 16, 64, 64])]; + tensor pair_15_cast_fp16 = reshape(shape = var_943, x = var_942_cast_fp16)[name = string("pair_15_cast_fp16")]; + tensor var_945_cast_fp16 = exp(x = pair_15_cast_fp16)[name = string("op_945_cast_fp16")]; + tensor pair_decay_3_cast_fp16 = mul(x = var_945_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_3_cast_fp16")]; + tensor k_beta_3_cast_fp16 = mul(x = k_15_cast_fp16, y = beta_7_cast_fp16)[name = string("k_beta_3_cast_fp16")]; + tensor v_beta_3_cast_fp16 = mul(x = v_7_cast_fp16, y = beta_7_cast_fp16)[name = string("v_beta_3_cast_fp16")]; + bool var_950_transpose_x_1 = const()[name = string("op_950_transpose_x_1"), val = bool(false)]; + bool var_950_transpose_y_1 = const()[name = string("op_950_transpose_y_1"), val = bool(true)]; + tensor var_950_cast_fp16 = matmul(transpose_x = var_950_transpose_x_1, transpose_y = var_950_transpose_y_1, x = k_beta_3_cast_fp16, y = k_15_cast_fp16)[name = string("op_950_cast_fp16")]; + tensor kkt_3_cast_fp16 = mul(x = var_950_cast_fp16, y = pair_decay_3_cast_fp16)[name = string("kkt_3_cast_fp16")]; + bool var_953_transpose_x_1 = const()[name = string("op_953_transpose_x_1"), val = bool(false)]; + bool var_953_transpose_y_1 = const()[name = string("op_953_transpose_y_1"), val = bool(true)]; + tensor var_953_cast_fp16 = matmul(transpose_x = var_953_transpose_x_1, transpose_y = var_953_transpose_y_1, x = q_19_cast_fp16, y = k_15_cast_fp16)[name = string("op_953_cast_fp16")]; + tensor attn_intra_3_cast_fp16 = mul(x = var_953_cast_fp16, y = pair_decay_3_cast_fp16)[name = string("attn_intra_3_cast_fp16")]; + tensor var_955_cast_fp16 = mul(x = kkt_3_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_955_cast_fp16")]; + tensor t_5_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_955_cast_fp16)[name = string("t_5_cast_fp16")]; + tensor var_959 = const()[name = string("op_959"), val = tensor([-1, 64, 64])]; + tensor t_7_cast_fp16 = reshape(shape = var_959, x = t_5_cast_fp16)[name = string("t_7_cast_fp16")]; + tensor var_964 = const()[name = string("op_964"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_965_cast_fp16 = reshape(shape = var_964, x = t_7_cast_fp16)[name = string("op_965_cast_fp16")]; + tensor var_970_cast_fp16 = mul(x = var_965_cast_fp16, y = var_216_to_fp16)[name = string("op_970_cast_fp16")]; + tensor blocks_13_axes_0 = const()[name = string("blocks_13_axes_0"), val = tensor([-2])]; + bool blocks_13_keep_dims_0 = const()[name = string("blocks_13_keep_dims_0"), val = bool(false)]; + tensor blocks_13_cast_fp16 = reduce_sum(axes = blocks_13_axes_0, keep_dims = blocks_13_keep_dims_0, x = var_970_cast_fp16)[name = string("blocks_13_cast_fp16")]; + tensor var_973_begin_0 = const()[name = string("op_973_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_973_end_0 = const()[name = string("op_973_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_973_end_mask_0 = const()[name = string("op_973_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_973_cast_fp16 = slice_by_index(begin = var_973_begin_0, end = var_973_end_0, end_mask = var_973_end_mask_0, x = blocks_13_cast_fp16)[name = string("op_973_cast_fp16")]; + tensor x_31_begin_0 = const()[name = string("x_31_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_31_end_0 = const()[name = string("x_31_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_31_end_mask_0 = const()[name = string("x_31_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_31_cast_fp16 = slice_by_index(begin = x_31_begin_0, end = x_31_end_0, end_mask = x_31_end_mask_0, x = var_973_cast_fp16)[name = string("x_31_cast_fp16")]; + bool var_983_transpose_x_0 = const()[name = string("op_983_transpose_x_0"), val = bool(false)]; + bool var_983_transpose_y_0 = const()[name = string("op_983_transpose_y_0"), val = bool(false)]; + tensor var_983_cast_fp16 = matmul(transpose_x = var_983_transpose_x_0, transpose_y = var_983_transpose_y_0, x = x_31_cast_fp16, y = var_224_to_fp16)[name = string("op_983_cast_fp16")]; + bool var_984_transpose_x_0 = const()[name = string("op_984_transpose_x_0"), val = bool(false)]; + bool var_984_transpose_y_0 = const()[name = string("op_984_transpose_y_0"), val = bool(false)]; + tensor var_984_cast_fp16 = matmul(transpose_x = var_984_transpose_x_0, transpose_y = var_984_transpose_y_0, x = var_224_to_fp16, y = var_983_cast_fp16)[name = string("op_984_cast_fp16")]; + fp16 const_18_promoted_to_fp16 = const()[name = string("const_18_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_13_cast_fp16 = mul(x = var_984_cast_fp16, y = const_18_promoted_to_fp16)[name = string("lower_13_cast_fp16")]; + bool bottom_13_interleave_0 = const()[name = string("bottom_13_interleave_0"), val = bool(false)]; + tensor bottom_13_cast_fp16 = concat(axis = var_811, interleave = bottom_13_interleave_0, values = (lower_13_cast_fp16, var_224_to_fp16))[name = string("bottom_13_cast_fp16")]; + bool inv_19_interleave_0 = const()[name = string("inv_19_interleave_0"), val = bool(false)]; + tensor inv_19_cast_fp16 = concat(axis = var_790, interleave = inv_19_interleave_0, values = (top_1_to_fp16, bottom_13_cast_fp16))[name = string("inv_19_cast_fp16")]; + tensor var_993 = const()[name = string("op_993"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_994_cast_fp16 = reshape(shape = var_993, x = t_7_cast_fp16)[name = string("op_994_cast_fp16")]; + tensor var_999_cast_fp16 = mul(x = var_994_cast_fp16, y = var_245_to_fp16)[name = string("op_999_cast_fp16")]; + tensor blocks_15_axes_0 = const()[name = string("blocks_15_axes_0"), val = tensor([-2])]; + bool blocks_15_keep_dims_0 = const()[name = string("blocks_15_keep_dims_0"), val = bool(false)]; + tensor blocks_15_cast_fp16 = reduce_sum(axes = blocks_15_axes_0, keep_dims = blocks_15_keep_dims_0, x = var_999_cast_fp16)[name = string("blocks_15_cast_fp16")]; + tensor var_1002_begin_0 = const()[name = string("op_1002_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_1002_end_0 = const()[name = string("op_1002_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_1002_end_mask_0 = const()[name = string("op_1002_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1002_cast_fp16 = slice_by_index(begin = var_1002_begin_0, end = var_1002_end_0, end_mask = var_1002_end_mask_0, x = blocks_15_cast_fp16)[name = string("op_1002_cast_fp16")]; + tensor x_33_begin_0 = const()[name = string("x_33_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_33_end_0 = const()[name = string("x_33_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_33_end_mask_0 = const()[name = string("x_33_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_33_cast_fp16 = slice_by_index(begin = x_33_begin_0, end = x_33_end_0, end_mask = x_33_end_mask_0, x = var_1002_cast_fp16)[name = string("x_33_cast_fp16")]; + tensor var_1004 = const()[name = string("op_1004"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_19_cast_fp16 = reshape(shape = var_1004, x = inv_19_cast_fp16)[name = string("pair_19_cast_fp16")]; + tensor var_1006_begin_0 = const()[name = string("op_1006_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1006_end_0 = const()[name = string("op_1006_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_1006_end_mask_0 = const()[name = string("op_1006_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1006_squeeze_mask_0 = const()[name = string("op_1006_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1006_cast_fp16 = slice_by_index(begin = var_1006_begin_0, end = var_1006_end_0, end_mask = var_1006_end_mask_0, squeeze_mask = var_1006_squeeze_mask_0, x = pair_19_cast_fp16)[name = string("op_1006_cast_fp16")]; + tensor var_1009_begin_0 = const()[name = string("op_1009_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1009_end_0 = const()[name = string("op_1009_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_1009_end_mask_0 = const()[name = string("op_1009_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1009_squeeze_mask_0 = const()[name = string("op_1009_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1009_cast_fp16 = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, squeeze_mask = var_1009_squeeze_mask_0, x = pair_19_cast_fp16)[name = string("op_1009_cast_fp16")]; + bool var_1012_transpose_x_0 = const()[name = string("op_1012_transpose_x_0"), val = bool(false)]; + bool var_1012_transpose_y_0 = const()[name = string("op_1012_transpose_y_0"), val = bool(false)]; + tensor var_1012_cast_fp16 = matmul(transpose_x = var_1012_transpose_x_0, transpose_y = var_1012_transpose_y_0, x = x_33_cast_fp16, y = var_1006_cast_fp16)[name = string("op_1012_cast_fp16")]; + bool var_1013_transpose_x_0 = const()[name = string("op_1013_transpose_x_0"), val = bool(false)]; + bool var_1013_transpose_y_0 = const()[name = string("op_1013_transpose_y_0"), val = bool(false)]; + tensor var_1013_cast_fp16 = matmul(transpose_x = var_1013_transpose_x_0, transpose_y = var_1013_transpose_y_0, x = var_1009_cast_fp16, y = var_1012_cast_fp16)[name = string("op_1013_cast_fp16")]; + fp16 const_19_promoted_to_fp16 = const()[name = string("const_19_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_15_cast_fp16 = mul(x = var_1013_cast_fp16, y = const_19_promoted_to_fp16)[name = string("lower_15_cast_fp16")]; + bool top_15_interleave_0 = const()[name = string("top_15_interleave_0"), val = bool(false)]; + tensor top_15_cast_fp16 = concat(axis = var_811, interleave = top_15_interleave_0, values = (var_1006_cast_fp16, var_262_to_fp16))[name = string("top_15_cast_fp16")]; + bool bottom_15_interleave_0 = const()[name = string("bottom_15_interleave_0"), val = bool(false)]; + tensor bottom_15_cast_fp16 = concat(axis = var_811, interleave = bottom_15_interleave_0, values = (lower_15_cast_fp16, var_1009_cast_fp16))[name = string("bottom_15_cast_fp16")]; + bool inv_21_interleave_0 = const()[name = string("inv_21_interleave_0"), val = bool(false)]; + tensor inv_21_cast_fp16 = concat(axis = var_790, interleave = inv_21_interleave_0, values = (top_15_cast_fp16, bottom_15_cast_fp16))[name = string("inv_21_cast_fp16")]; + tensor var_1022 = const()[name = string("op_1022"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_1023_cast_fp16 = reshape(shape = var_1022, x = t_7_cast_fp16)[name = string("op_1023_cast_fp16")]; + tensor var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292950208)))]; + tensor var_1028_cast_fp16 = mul(x = var_1023_cast_fp16, y = var_1027_to_fp16)[name = string("op_1028_cast_fp16")]; + tensor blocks_17_axes_0 = const()[name = string("blocks_17_axes_0"), val = tensor([-2])]; + bool blocks_17_keep_dims_0 = const()[name = string("blocks_17_keep_dims_0"), val = bool(false)]; + tensor blocks_17_cast_fp16 = reduce_sum(axes = blocks_17_axes_0, keep_dims = blocks_17_keep_dims_0, x = var_1028_cast_fp16)[name = string("blocks_17_cast_fp16")]; + tensor var_1031_begin_0 = const()[name = string("op_1031_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_1031_end_0 = const()[name = string("op_1031_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_1031_end_mask_0 = const()[name = string("op_1031_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1031_cast_fp16 = slice_by_index(begin = var_1031_begin_0, end = var_1031_end_0, end_mask = var_1031_end_mask_0, x = blocks_17_cast_fp16)[name = string("op_1031_cast_fp16")]; + tensor x_35_begin_0 = const()[name = string("x_35_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_35_end_0 = const()[name = string("x_35_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_35_end_mask_0 = const()[name = string("x_35_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_35_cast_fp16 = slice_by_index(begin = x_35_begin_0, end = x_35_end_0, end_mask = x_35_end_mask_0, x = var_1031_cast_fp16)[name = string("x_35_cast_fp16")]; + tensor var_1033 = const()[name = string("op_1033"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_21_cast_fp16 = reshape(shape = var_1033, x = inv_21_cast_fp16)[name = string("pair_21_cast_fp16")]; + tensor var_1035_begin_0 = const()[name = string("op_1035_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1035_end_0 = const()[name = string("op_1035_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_1035_end_mask_0 = const()[name = string("op_1035_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1035_squeeze_mask_0 = const()[name = string("op_1035_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1035_cast_fp16 = slice_by_index(begin = var_1035_begin_0, end = var_1035_end_0, end_mask = var_1035_end_mask_0, squeeze_mask = var_1035_squeeze_mask_0, x = pair_21_cast_fp16)[name = string("op_1035_cast_fp16")]; + tensor var_1038_begin_0 = const()[name = string("op_1038_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1038_end_0 = const()[name = string("op_1038_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_1038_end_mask_0 = const()[name = string("op_1038_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1038_squeeze_mask_0 = const()[name = string("op_1038_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1038_cast_fp16 = slice_by_index(begin = var_1038_begin_0, end = var_1038_end_0, end_mask = var_1038_end_mask_0, squeeze_mask = var_1038_squeeze_mask_0, x = pair_21_cast_fp16)[name = string("op_1038_cast_fp16")]; + bool var_1041_transpose_x_0 = const()[name = string("op_1041_transpose_x_0"), val = bool(false)]; + bool var_1041_transpose_y_0 = const()[name = string("op_1041_transpose_y_0"), val = bool(false)]; + tensor var_1041_cast_fp16 = matmul(transpose_x = var_1041_transpose_x_0, transpose_y = var_1041_transpose_y_0, x = x_35_cast_fp16, y = var_1035_cast_fp16)[name = string("op_1041_cast_fp16")]; + bool var_1042_transpose_x_0 = const()[name = string("op_1042_transpose_x_0"), val = bool(false)]; + bool var_1042_transpose_y_0 = const()[name = string("op_1042_transpose_y_0"), val = bool(false)]; + tensor var_1042_cast_fp16 = matmul(transpose_x = var_1042_transpose_x_0, transpose_y = var_1042_transpose_y_0, x = var_1038_cast_fp16, y = var_1041_cast_fp16)[name = string("op_1042_cast_fp16")]; + fp16 const_20_promoted_to_fp16 = const()[name = string("const_20_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_17_cast_fp16 = mul(x = var_1042_cast_fp16, y = const_20_promoted_to_fp16)[name = string("lower_17_cast_fp16")]; + bool top_17_interleave_0 = const()[name = string("top_17_interleave_0"), val = bool(false)]; + tensor top_17_cast_fp16 = concat(axis = var_811, interleave = top_17_interleave_0, values = (var_1035_cast_fp16, var_291_to_fp16))[name = string("top_17_cast_fp16")]; + bool bottom_17_interleave_0 = const()[name = string("bottom_17_interleave_0"), val = bool(false)]; + tensor bottom_17_cast_fp16 = concat(axis = var_811, interleave = bottom_17_interleave_0, values = (lower_17_cast_fp16, var_1038_cast_fp16))[name = string("bottom_17_cast_fp16")]; + bool inv_23_interleave_0 = const()[name = string("inv_23_interleave_0"), val = bool(false)]; + tensor inv_23_cast_fp16 = concat(axis = var_790, interleave = inv_23_interleave_0, values = (top_17_cast_fp16, bottom_17_cast_fp16))[name = string("inv_23_cast_fp16")]; + tensor var_1051 = const()[name = string("op_1051"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_1052_cast_fp16 = reshape(shape = var_1051, x = t_7_cast_fp16)[name = string("op_1052_cast_fp16")]; + tensor var_1056_to_fp16 = const()[name = string("op_1056_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292950400)))]; + tensor var_1057_cast_fp16 = mul(x = var_1052_cast_fp16, y = var_1056_to_fp16)[name = string("op_1057_cast_fp16")]; + tensor blocks_19_axes_0 = const()[name = string("blocks_19_axes_0"), val = tensor([-2])]; + bool blocks_19_keep_dims_0 = const()[name = string("blocks_19_keep_dims_0"), val = bool(false)]; + tensor blocks_19_cast_fp16 = reduce_sum(axes = blocks_19_axes_0, keep_dims = blocks_19_keep_dims_0, x = var_1057_cast_fp16)[name = string("blocks_19_cast_fp16")]; + tensor var_1060_begin_0 = const()[name = string("op_1060_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_1060_end_0 = const()[name = string("op_1060_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_1060_end_mask_0 = const()[name = string("op_1060_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1060_cast_fp16 = slice_by_index(begin = var_1060_begin_0, end = var_1060_end_0, end_mask = var_1060_end_mask_0, x = blocks_19_cast_fp16)[name = string("op_1060_cast_fp16")]; + tensor x_37_begin_0 = const()[name = string("x_37_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_37_end_0 = const()[name = string("x_37_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_37_end_mask_0 = const()[name = string("x_37_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_37_cast_fp16 = slice_by_index(begin = x_37_begin_0, end = x_37_end_0, end_mask = x_37_end_mask_0, x = var_1060_cast_fp16)[name = string("x_37_cast_fp16")]; + tensor var_1062 = const()[name = string("op_1062"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_23_cast_fp16 = reshape(shape = var_1062, x = inv_23_cast_fp16)[name = string("pair_23_cast_fp16")]; + tensor var_1064_begin_0 = const()[name = string("op_1064_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1064_end_0 = const()[name = string("op_1064_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_1064_end_mask_0 = const()[name = string("op_1064_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1064_squeeze_mask_0 = const()[name = string("op_1064_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1064_cast_fp16 = slice_by_index(begin = var_1064_begin_0, end = var_1064_end_0, end_mask = var_1064_end_mask_0, squeeze_mask = var_1064_squeeze_mask_0, x = pair_23_cast_fp16)[name = string("op_1064_cast_fp16")]; + tensor var_1067_begin_0 = const()[name = string("op_1067_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1067_end_0 = const()[name = string("op_1067_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_1067_end_mask_0 = const()[name = string("op_1067_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1067_squeeze_mask_0 = const()[name = string("op_1067_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1067_cast_fp16 = slice_by_index(begin = var_1067_begin_0, end = var_1067_end_0, end_mask = var_1067_end_mask_0, squeeze_mask = var_1067_squeeze_mask_0, x = pair_23_cast_fp16)[name = string("op_1067_cast_fp16")]; + bool var_1070_transpose_x_0 = const()[name = string("op_1070_transpose_x_0"), val = bool(false)]; + bool var_1070_transpose_y_0 = const()[name = string("op_1070_transpose_y_0"), val = bool(false)]; + tensor var_1070_cast_fp16 = matmul(transpose_x = var_1070_transpose_x_0, transpose_y = var_1070_transpose_y_0, x = x_37_cast_fp16, y = var_1064_cast_fp16)[name = string("op_1070_cast_fp16")]; + bool var_1071_transpose_x_0 = const()[name = string("op_1071_transpose_x_0"), val = bool(false)]; + bool var_1071_transpose_y_0 = const()[name = string("op_1071_transpose_y_0"), val = bool(false)]; + tensor var_1071_cast_fp16 = matmul(transpose_x = var_1071_transpose_x_0, transpose_y = var_1071_transpose_y_0, x = var_1067_cast_fp16, y = var_1070_cast_fp16)[name = string("op_1071_cast_fp16")]; + fp16 const_21_promoted_to_fp16 = const()[name = string("const_21_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_19_cast_fp16 = mul(x = var_1071_cast_fp16, y = const_21_promoted_to_fp16)[name = string("lower_19_cast_fp16")]; + bool top_19_interleave_0 = const()[name = string("top_19_interleave_0"), val = bool(false)]; + tensor top_19_cast_fp16 = concat(axis = var_811, interleave = top_19_interleave_0, values = (var_1064_cast_fp16, var_320_to_fp16))[name = string("top_19_cast_fp16")]; + bool bottom_19_interleave_0 = const()[name = string("bottom_19_interleave_0"), val = bool(false)]; + tensor bottom_19_cast_fp16 = concat(axis = var_811, interleave = bottom_19_interleave_0, values = (lower_19_cast_fp16, var_1067_cast_fp16))[name = string("bottom_19_cast_fp16")]; + bool inv_25_interleave_0 = const()[name = string("inv_25_interleave_0"), val = bool(false)]; + tensor inv_25_cast_fp16 = concat(axis = var_790, interleave = inv_25_interleave_0, values = (top_19_cast_fp16, bottom_19_cast_fp16))[name = string("inv_25_cast_fp16")]; + tensor var_1080 = const()[name = string("op_1080"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_1081_cast_fp16 = reshape(shape = var_1080, x = t_7_cast_fp16)[name = string("op_1081_cast_fp16")]; + tensor var_1085_to_fp16 = const()[name = string("op_1085_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_1086_cast_fp16 = mul(x = var_1081_cast_fp16, y = var_1085_to_fp16)[name = string("op_1086_cast_fp16")]; + tensor blocks_21_axes_0 = const()[name = string("blocks_21_axes_0"), val = tensor([-2])]; + bool blocks_21_keep_dims_0 = const()[name = string("blocks_21_keep_dims_0"), val = bool(false)]; + tensor blocks_21_cast_fp16 = reduce_sum(axes = blocks_21_axes_0, keep_dims = blocks_21_keep_dims_0, x = var_1086_cast_fp16)[name = string("blocks_21_cast_fp16")]; + tensor var_1089_begin_0 = const()[name = string("op_1089_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_1089_end_0 = const()[name = string("op_1089_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_1089_end_mask_0 = const()[name = string("op_1089_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = blocks_21_cast_fp16)[name = string("op_1089_cast_fp16")]; + tensor x_39_begin_0 = const()[name = string("x_39_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_39_end_0 = const()[name = string("x_39_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_39_end_mask_0 = const()[name = string("x_39_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_39_cast_fp16 = slice_by_index(begin = x_39_begin_0, end = x_39_end_0, end_mask = x_39_end_mask_0, x = var_1089_cast_fp16)[name = string("x_39_cast_fp16")]; + tensor var_1091 = const()[name = string("op_1091"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_25_cast_fp16 = reshape(shape = var_1091, x = inv_25_cast_fp16)[name = string("pair_25_cast_fp16")]; + tensor var_1093_begin_0 = const()[name = string("op_1093_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1093_end_0 = const()[name = string("op_1093_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_1093_end_mask_0 = const()[name = string("op_1093_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1093_squeeze_mask_0 = const()[name = string("op_1093_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1093_cast_fp16 = slice_by_index(begin = var_1093_begin_0, end = var_1093_end_0, end_mask = var_1093_end_mask_0, squeeze_mask = var_1093_squeeze_mask_0, x = pair_25_cast_fp16)[name = string("op_1093_cast_fp16")]; + tensor var_1096_begin_0 = const()[name = string("op_1096_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1096_end_0 = const()[name = string("op_1096_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_1096_end_mask_0 = const()[name = string("op_1096_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1096_squeeze_mask_0 = const()[name = string("op_1096_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1096_cast_fp16 = slice_by_index(begin = var_1096_begin_0, end = var_1096_end_0, end_mask = var_1096_end_mask_0, squeeze_mask = var_1096_squeeze_mask_0, x = pair_25_cast_fp16)[name = string("op_1096_cast_fp16")]; + bool var_1099_transpose_x_0 = const()[name = string("op_1099_transpose_x_0"), val = bool(false)]; + bool var_1099_transpose_y_0 = const()[name = string("op_1099_transpose_y_0"), val = bool(false)]; + tensor var_1099_cast_fp16 = matmul(transpose_x = var_1099_transpose_x_0, transpose_y = var_1099_transpose_y_0, x = x_39_cast_fp16, y = var_1093_cast_fp16)[name = string("op_1099_cast_fp16")]; + bool var_1100_transpose_x_0 = const()[name = string("op_1100_transpose_x_0"), val = bool(false)]; + bool var_1100_transpose_y_0 = const()[name = string("op_1100_transpose_y_0"), val = bool(false)]; + tensor var_1100_cast_fp16 = matmul(transpose_x = var_1100_transpose_x_0, transpose_y = var_1100_transpose_y_0, x = var_1096_cast_fp16, y = var_1099_cast_fp16)[name = string("op_1100_cast_fp16")]; + fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_21_cast_fp16 = mul(x = var_1100_cast_fp16, y = const_22_promoted_to_fp16)[name = string("lower_21_cast_fp16")]; + bool top_21_interleave_0 = const()[name = string("top_21_interleave_0"), val = bool(false)]; + tensor top_21_cast_fp16 = concat(axis = var_811, interleave = top_21_interleave_0, values = (var_1093_cast_fp16, var_349_to_fp16))[name = string("top_21_cast_fp16")]; + bool bottom_21_interleave_0 = const()[name = string("bottom_21_interleave_0"), val = bool(false)]; + tensor bottom_21_cast_fp16 = concat(axis = var_811, interleave = bottom_21_interleave_0, values = (lower_21_cast_fp16, var_1096_cast_fp16))[name = string("bottom_21_cast_fp16")]; + bool inv_27_interleave_0 = const()[name = string("inv_27_interleave_0"), val = bool(false)]; + tensor inv_27_cast_fp16 = concat(axis = var_790, interleave = inv_27_interleave_0, values = (top_21_cast_fp16, bottom_21_cast_fp16))[name = string("inv_27_cast_fp16")]; + tensor var_1109 = const()[name = string("op_1109"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_1110_cast_fp16 = reshape(shape = var_1109, x = t_7_cast_fp16)[name = string("op_1110_cast_fp16")]; + tensor blocks_23_axes_0 = const()[name = string("blocks_23_axes_0"), val = tensor([-2])]; + bool blocks_23_keep_dims_0 = const()[name = string("blocks_23_keep_dims_0"), val = bool(false)]; + tensor blocks_23_cast_fp16 = reduce_sum(axes = blocks_23_axes_0, keep_dims = blocks_23_keep_dims_0, x = var_1110_cast_fp16)[name = string("blocks_23_cast_fp16")]; + tensor var_1118_begin_0 = const()[name = string("op_1118_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_1118_end_0 = const()[name = string("op_1118_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_1118_end_mask_0 = const()[name = string("op_1118_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1118_cast_fp16 = slice_by_index(begin = var_1118_begin_0, end = var_1118_end_0, end_mask = var_1118_end_mask_0, x = blocks_23_cast_fp16)[name = string("op_1118_cast_fp16")]; + tensor x_41_begin_0 = const()[name = string("x_41_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_41_end_0 = const()[name = string("x_41_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_41_end_mask_0 = const()[name = string("x_41_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_41_cast_fp16 = slice_by_index(begin = x_41_begin_0, end = x_41_end_0, end_mask = x_41_end_mask_0, x = var_1118_cast_fp16)[name = string("x_41_cast_fp16")]; + tensor var_1120 = const()[name = string("op_1120"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_27_cast_fp16 = reshape(shape = var_1120, x = inv_27_cast_fp16)[name = string("pair_27_cast_fp16")]; + tensor var_1122_begin_0 = const()[name = string("op_1122_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1122_end_0 = const()[name = string("op_1122_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_1122_end_mask_0 = const()[name = string("op_1122_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1122_squeeze_mask_0 = const()[name = string("op_1122_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1122_cast_fp16 = slice_by_index(begin = var_1122_begin_0, end = var_1122_end_0, end_mask = var_1122_end_mask_0, squeeze_mask = var_1122_squeeze_mask_0, x = pair_27_cast_fp16)[name = string("op_1122_cast_fp16")]; + tensor var_1125_begin_0 = const()[name = string("op_1125_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1125_end_0 = const()[name = string("op_1125_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_1125_end_mask_0 = const()[name = string("op_1125_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1125_squeeze_mask_0 = const()[name = string("op_1125_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1125_cast_fp16 = slice_by_index(begin = var_1125_begin_0, end = var_1125_end_0, end_mask = var_1125_end_mask_0, squeeze_mask = var_1125_squeeze_mask_0, x = pair_27_cast_fp16)[name = string("op_1125_cast_fp16")]; + bool var_1128_transpose_x_0 = const()[name = string("op_1128_transpose_x_0"), val = bool(false)]; + bool var_1128_transpose_y_0 = const()[name = string("op_1128_transpose_y_0"), val = bool(false)]; + tensor var_1128_cast_fp16 = matmul(transpose_x = var_1128_transpose_x_0, transpose_y = var_1128_transpose_y_0, x = x_41_cast_fp16, y = var_1122_cast_fp16)[name = string("op_1128_cast_fp16")]; + bool var_1129_transpose_x_0 = const()[name = string("op_1129_transpose_x_0"), val = bool(false)]; + bool var_1129_transpose_y_0 = const()[name = string("op_1129_transpose_y_0"), val = bool(false)]; + tensor var_1129_cast_fp16 = matmul(transpose_x = var_1129_transpose_x_0, transpose_y = var_1129_transpose_y_0, x = var_1125_cast_fp16, y = var_1128_cast_fp16)[name = string("op_1129_cast_fp16")]; + fp16 const_23_promoted_to_fp16 = const()[name = string("const_23_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_23_cast_fp16 = mul(x = var_1129_cast_fp16, y = const_23_promoted_to_fp16)[name = string("lower_23_cast_fp16")]; + bool top_23_interleave_0 = const()[name = string("top_23_interleave_0"), val = bool(false)]; + tensor top_23_cast_fp16 = concat(axis = var_811, interleave = top_23_interleave_0, values = (var_1122_cast_fp16, var_378_to_fp16))[name = string("top_23_cast_fp16")]; + bool bottom_23_interleave_0 = const()[name = string("bottom_23_interleave_0"), val = bool(false)]; + tensor bottom_23_cast_fp16 = concat(axis = var_811, interleave = bottom_23_interleave_0, values = (lower_23_cast_fp16, var_1125_cast_fp16))[name = string("bottom_23_cast_fp16")]; + bool inv_29_interleave_0 = const()[name = string("inv_29_interleave_0"), val = bool(false)]; + tensor inv_29_cast_fp16 = concat(axis = var_790, interleave = inv_29_interleave_0, values = (top_23_cast_fp16, bottom_23_cast_fp16))[name = string("inv_29_cast_fp16")]; + tensor var_1138 = const()[name = string("op_1138"), val = tensor([32, 16, 64, 64])]; + tensor inv_31_cast_fp16 = reshape(shape = var_1138, x = inv_29_cast_fp16)[name = string("inv_31_cast_fp16")]; + bool new_v_3_transpose_x_0 = const()[name = string("new_v_3_transpose_x_0"), val = bool(false)]; + bool new_v_3_transpose_y_0 = const()[name = string("new_v_3_transpose_y_0"), val = bool(false)]; + tensor new_v_3_cast_fp16 = matmul(transpose_x = new_v_3_transpose_x_0, transpose_y = new_v_3_transpose_y_0, x = inv_31_cast_fp16, y = v_beta_3_cast_fp16)[name = string("new_v_3_cast_fp16")]; + tensor var_1141_cast_fp16 = exp(x = cum_3_cast_fp16)[name = string("op_1141_cast_fp16")]; + tensor var_1142_axes_0 = const()[name = string("op_1142_axes_0"), val = tensor([3])]; + tensor var_1142_cast_fp16 = expand_dims(axes = var_1142_axes_0, x = var_1141_cast_fp16)[name = string("op_1142_cast_fp16")]; + tensor var_1143_cast_fp16 = mul(x = k_beta_3_cast_fp16, y = var_1142_cast_fp16)[name = string("op_1143_cast_fp16")]; + bool k_cumdecay_3_transpose_x_0 = const()[name = string("k_cumdecay_3_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_3_transpose_y_0 = const()[name = string("k_cumdecay_3_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_3_cast_fp16 = matmul(transpose_x = k_cumdecay_3_transpose_x_0, transpose_y = k_cumdecay_3_transpose_y_0, x = inv_31_cast_fp16, y = var_1143_cast_fp16)[name = string("k_cumdecay_3_cast_fp16")]; + tensor q_dec_3_cast_fp16 = mul(x = q_19_cast_fp16, y = var_1142_cast_fp16)[name = string("q_dec_3_cast_fp16")]; + tensor var_1148_cast_fp16 = exp(x = to_end_3_cast_fp16)[name = string("op_1148_cast_fp16")]; + tensor var_1149_axes_0 = const()[name = string("op_1149_axes_0"), val = tensor([3])]; + tensor var_1149_cast_fp16 = expand_dims(axes = var_1149_axes_0, x = var_1148_cast_fp16)[name = string("op_1149_cast_fp16")]; + tensor k_dec_3_cast_fp16 = mul(x = k_15_cast_fp16, y = var_1149_cast_fp16)[name = string("k_dec_3_cast_fp16")]; + tensor var_1151_begin_0 = const()[name = string("op_1151_begin_0"), val = tensor([0, 0, -1])]; + tensor var_1151_end_0 = const()[name = string("op_1151_end_0"), val = tensor([32, 16, 64])]; + tensor var_1151_end_mask_0 = const()[name = string("op_1151_end_mask_0"), val = tensor([true, true, true])]; + tensor var_1151_cast_fp16 = slice_by_index(begin = var_1151_begin_0, end = var_1151_end_0, end_mask = var_1151_end_mask_0, x = cum_3_cast_fp16)[name = string("op_1151_cast_fp16")]; + tensor var_1152_cast_fp16 = exp(x = var_1151_cast_fp16)[name = string("op_1152_cast_fp16")]; + tensor chunk_decay_3_axes_0 = const()[name = string("chunk_decay_3_axes_0"), val = tensor([3])]; + tensor chunk_decay_3_cast_fp16 = expand_dims(axes = chunk_decay_3_axes_0, x = var_1152_cast_fp16)[name = string("chunk_decay_3_cast_fp16")]; + tensor var_1157_begin_0 = const()[name = string("op_1157_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1157_end_0 = const()[name = string("op_1157_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1157_end_mask_0 = const()[name = string("op_1157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1157_squeeze_mask_0 = const()[name = string("op_1157_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1157_cast_fp16 = slice_by_index(begin = var_1157_begin_0, end = var_1157_end_0, end_mask = var_1157_end_mask_0, squeeze_mask = var_1157_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1157_cast_fp16")]; + tensor var_1159_begin_0 = const()[name = string("op_1159_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1159_end_0 = const()[name = string("op_1159_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1159_end_mask_0 = const()[name = string("op_1159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1159_squeeze_mask_0 = const()[name = string("op_1159_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1159_cast_fp16 = slice_by_index(begin = var_1159_begin_0, end = var_1159_end_0, end_mask = var_1159_end_mask_0, squeeze_mask = var_1159_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1159_cast_fp16")]; + bool var_1160_transpose_x_0 = const()[name = string("op_1160_transpose_x_0"), val = bool(false)]; + bool var_1160_transpose_y_0 = const()[name = string("op_1160_transpose_y_0"), val = bool(false)]; + tensor var_1160_cast_fp16 = matmul(transpose_x = var_1160_transpose_x_0, transpose_y = var_1160_transpose_y_0, x = var_1159_cast_fp16, y = state_1_to_fp16)[name = string("op_1160_cast_fp16")]; + tensor v_new_33_cast_fp16 = sub(x = var_1157_cast_fp16, y = var_1160_cast_fp16)[name = string("v_new_33_cast_fp16")]; + tensor var_1163_begin_0 = const()[name = string("op_1163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1163_end_0 = const()[name = string("op_1163_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1163_end_mask_0 = const()[name = string("op_1163_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1163_squeeze_mask_0 = const()[name = string("op_1163_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1163_cast_fp16 = slice_by_index(begin = var_1163_begin_0, end = var_1163_end_0, end_mask = var_1163_end_mask_0, squeeze_mask = var_1163_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1163_cast_fp16")]; + bool var_1164_transpose_x_0 = const()[name = string("op_1164_transpose_x_0"), val = bool(false)]; + bool var_1164_transpose_y_0 = const()[name = string("op_1164_transpose_y_0"), val = bool(false)]; + tensor var_1164_cast_fp16 = matmul(transpose_x = var_1164_transpose_x_0, transpose_y = var_1164_transpose_y_0, x = var_1163_cast_fp16, y = state_1_to_fp16)[name = string("op_1164_cast_fp16")]; + tensor var_1166_begin_0 = const()[name = string("op_1166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1166_end_0 = const()[name = string("op_1166_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_1166_end_mask_0 = const()[name = string("op_1166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1166_squeeze_mask_0 = const()[name = string("op_1166_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1166_cast_fp16 = slice_by_index(begin = var_1166_begin_0, end = var_1166_end_0, end_mask = var_1166_end_mask_0, squeeze_mask = var_1166_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1166_cast_fp16")]; + bool var_1167_transpose_x_0 = const()[name = string("op_1167_transpose_x_0"), val = bool(false)]; + bool var_1167_transpose_y_0 = const()[name = string("op_1167_transpose_y_0"), val = bool(false)]; + tensor var_1167_cast_fp16 = matmul(transpose_x = var_1167_transpose_x_0, transpose_y = var_1167_transpose_y_0, x = var_1166_cast_fp16, y = v_new_33_cast_fp16)[name = string("op_1167_cast_fp16")]; + tensor var_1168_cast_fp16 = add(x = var_1164_cast_fp16, y = var_1167_cast_fp16)[name = string("op_1168_cast_fp16")]; + tensor var_1170_begin_0 = const()[name = string("op_1170_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1170_end_0 = const()[name = string("op_1170_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_1170_end_mask_0 = const()[name = string("op_1170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1170_squeeze_mask_0 = const()[name = string("op_1170_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1170_cast_fp16 = slice_by_index(begin = var_1170_begin_0, end = var_1170_end_0, end_mask = var_1170_end_mask_0, squeeze_mask = var_1170_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1170_cast_fp16")]; + tensor var_1171_cast_fp16 = mul(x = state_1_to_fp16, y = var_1170_cast_fp16)[name = string("op_1171_cast_fp16")]; + tensor var_1173_begin_0 = const()[name = string("op_1173_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1173_end_0 = const()[name = string("op_1173_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1173_end_mask_0 = const()[name = string("op_1173_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1173_squeeze_mask_0 = const()[name = string("op_1173_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1173_cast_fp16 = slice_by_index(begin = var_1173_begin_0, end = var_1173_end_0, end_mask = var_1173_end_mask_0, squeeze_mask = var_1173_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1173_cast_fp16")]; + bool var_1175_transpose_x_1 = const()[name = string("op_1175_transpose_x_1"), val = bool(true)]; + bool var_1175_transpose_y_1 = const()[name = string("op_1175_transpose_y_1"), val = bool(false)]; + tensor var_1175_cast_fp16 = matmul(transpose_x = var_1175_transpose_x_1, transpose_y = var_1175_transpose_y_1, x = var_1173_cast_fp16, y = v_new_33_cast_fp16)[name = string("op_1175_cast_fp16")]; + tensor state_35_cast_fp16 = add(x = var_1171_cast_fp16, y = var_1175_cast_fp16)[name = string("state_35_cast_fp16")]; + tensor var_1178_begin_0 = const()[name = string("op_1178_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1178_end_0 = const()[name = string("op_1178_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1178_end_mask_0 = const()[name = string("op_1178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1178_squeeze_mask_0 = const()[name = string("op_1178_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1178_cast_fp16 = slice_by_index(begin = var_1178_begin_0, end = var_1178_end_0, end_mask = var_1178_end_mask_0, squeeze_mask = var_1178_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1178_cast_fp16")]; + tensor var_1180_begin_0 = const()[name = string("op_1180_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1180_end_0 = const()[name = string("op_1180_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1180_end_mask_0 = const()[name = string("op_1180_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1180_squeeze_mask_0 = const()[name = string("op_1180_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1180_cast_fp16 = slice_by_index(begin = var_1180_begin_0, end = var_1180_end_0, end_mask = var_1180_end_mask_0, squeeze_mask = var_1180_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1180_cast_fp16")]; + bool var_1181_transpose_x_0 = const()[name = string("op_1181_transpose_x_0"), val = bool(false)]; + bool var_1181_transpose_y_0 = const()[name = string("op_1181_transpose_y_0"), val = bool(false)]; + tensor var_1181_cast_fp16 = matmul(transpose_x = var_1181_transpose_x_0, transpose_y = var_1181_transpose_y_0, x = var_1180_cast_fp16, y = state_35_cast_fp16)[name = string("op_1181_cast_fp16")]; + tensor v_new_35_cast_fp16 = sub(x = var_1178_cast_fp16, y = var_1181_cast_fp16)[name = string("v_new_35_cast_fp16")]; + tensor var_1184_begin_0 = const()[name = string("op_1184_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1184_end_0 = const()[name = string("op_1184_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1184_end_mask_0 = const()[name = string("op_1184_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1184_squeeze_mask_0 = const()[name = string("op_1184_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1184_cast_fp16 = slice_by_index(begin = var_1184_begin_0, end = var_1184_end_0, end_mask = var_1184_end_mask_0, squeeze_mask = var_1184_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1184_cast_fp16")]; + bool var_1185_transpose_x_0 = const()[name = string("op_1185_transpose_x_0"), val = bool(false)]; + bool var_1185_transpose_y_0 = const()[name = string("op_1185_transpose_y_0"), val = bool(false)]; + tensor var_1185_cast_fp16 = matmul(transpose_x = var_1185_transpose_x_0, transpose_y = var_1185_transpose_y_0, x = var_1184_cast_fp16, y = state_35_cast_fp16)[name = string("op_1185_cast_fp16")]; + tensor var_1187_begin_0 = const()[name = string("op_1187_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1187_end_0 = const()[name = string("op_1187_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_1187_end_mask_0 = const()[name = string("op_1187_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1187_squeeze_mask_0 = const()[name = string("op_1187_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1187_cast_fp16 = slice_by_index(begin = var_1187_begin_0, end = var_1187_end_0, end_mask = var_1187_end_mask_0, squeeze_mask = var_1187_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1187_cast_fp16")]; + bool var_1188_transpose_x_0 = const()[name = string("op_1188_transpose_x_0"), val = bool(false)]; + bool var_1188_transpose_y_0 = const()[name = string("op_1188_transpose_y_0"), val = bool(false)]; + tensor var_1188_cast_fp16 = matmul(transpose_x = var_1188_transpose_x_0, transpose_y = var_1188_transpose_y_0, x = var_1187_cast_fp16, y = v_new_35_cast_fp16)[name = string("op_1188_cast_fp16")]; + tensor var_1189_cast_fp16 = add(x = var_1185_cast_fp16, y = var_1188_cast_fp16)[name = string("op_1189_cast_fp16")]; + tensor var_1191_begin_0 = const()[name = string("op_1191_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1191_end_0 = const()[name = string("op_1191_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_1191_end_mask_0 = const()[name = string("op_1191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1191_squeeze_mask_0 = const()[name = string("op_1191_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1191_cast_fp16 = slice_by_index(begin = var_1191_begin_0, end = var_1191_end_0, end_mask = var_1191_end_mask_0, squeeze_mask = var_1191_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1191_cast_fp16")]; + tensor var_1192_cast_fp16 = mul(x = state_35_cast_fp16, y = var_1191_cast_fp16)[name = string("op_1192_cast_fp16")]; + tensor var_1194_begin_0 = const()[name = string("op_1194_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1194_end_0 = const()[name = string("op_1194_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1194_end_mask_0 = const()[name = string("op_1194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1194_squeeze_mask_0 = const()[name = string("op_1194_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, squeeze_mask = var_1194_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1194_cast_fp16")]; + bool var_1196_transpose_x_1 = const()[name = string("op_1196_transpose_x_1"), val = bool(true)]; + bool var_1196_transpose_y_1 = const()[name = string("op_1196_transpose_y_1"), val = bool(false)]; + tensor var_1196_cast_fp16 = matmul(transpose_x = var_1196_transpose_x_1, transpose_y = var_1196_transpose_y_1, x = var_1194_cast_fp16, y = v_new_35_cast_fp16)[name = string("op_1196_cast_fp16")]; + tensor state_37_cast_fp16 = add(x = var_1192_cast_fp16, y = var_1196_cast_fp16)[name = string("state_37_cast_fp16")]; + tensor var_1199_begin_0 = const()[name = string("op_1199_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1199_end_0 = const()[name = string("op_1199_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1199_end_mask_0 = const()[name = string("op_1199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1199_squeeze_mask_0 = const()[name = string("op_1199_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1199_cast_fp16 = slice_by_index(begin = var_1199_begin_0, end = var_1199_end_0, end_mask = var_1199_end_mask_0, squeeze_mask = var_1199_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1199_cast_fp16")]; + tensor var_1201_begin_0 = const()[name = string("op_1201_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1201_end_0 = const()[name = string("op_1201_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1201_end_mask_0 = const()[name = string("op_1201_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1201_squeeze_mask_0 = const()[name = string("op_1201_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1201_cast_fp16 = slice_by_index(begin = var_1201_begin_0, end = var_1201_end_0, end_mask = var_1201_end_mask_0, squeeze_mask = var_1201_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1201_cast_fp16")]; + bool var_1202_transpose_x_0 = const()[name = string("op_1202_transpose_x_0"), val = bool(false)]; + bool var_1202_transpose_y_0 = const()[name = string("op_1202_transpose_y_0"), val = bool(false)]; + tensor var_1202_cast_fp16 = matmul(transpose_x = var_1202_transpose_x_0, transpose_y = var_1202_transpose_y_0, x = var_1201_cast_fp16, y = state_37_cast_fp16)[name = string("op_1202_cast_fp16")]; + tensor v_new_37_cast_fp16 = sub(x = var_1199_cast_fp16, y = var_1202_cast_fp16)[name = string("v_new_37_cast_fp16")]; + tensor var_1205_begin_0 = const()[name = string("op_1205_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1205_end_0 = const()[name = string("op_1205_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1205_end_mask_0 = const()[name = string("op_1205_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1205_squeeze_mask_0 = const()[name = string("op_1205_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1205_cast_fp16 = slice_by_index(begin = var_1205_begin_0, end = var_1205_end_0, end_mask = var_1205_end_mask_0, squeeze_mask = var_1205_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1205_cast_fp16")]; + bool var_1206_transpose_x_0 = const()[name = string("op_1206_transpose_x_0"), val = bool(false)]; + bool var_1206_transpose_y_0 = const()[name = string("op_1206_transpose_y_0"), val = bool(false)]; + tensor var_1206_cast_fp16 = matmul(transpose_x = var_1206_transpose_x_0, transpose_y = var_1206_transpose_y_0, x = var_1205_cast_fp16, y = state_37_cast_fp16)[name = string("op_1206_cast_fp16")]; + tensor var_1208_begin_0 = const()[name = string("op_1208_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1208_end_0 = const()[name = string("op_1208_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_1208_end_mask_0 = const()[name = string("op_1208_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1208_squeeze_mask_0 = const()[name = string("op_1208_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1208_cast_fp16 = slice_by_index(begin = var_1208_begin_0, end = var_1208_end_0, end_mask = var_1208_end_mask_0, squeeze_mask = var_1208_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1208_cast_fp16")]; + bool var_1209_transpose_x_0 = const()[name = string("op_1209_transpose_x_0"), val = bool(false)]; + bool var_1209_transpose_y_0 = const()[name = string("op_1209_transpose_y_0"), val = bool(false)]; + tensor var_1209_cast_fp16 = matmul(transpose_x = var_1209_transpose_x_0, transpose_y = var_1209_transpose_y_0, x = var_1208_cast_fp16, y = v_new_37_cast_fp16)[name = string("op_1209_cast_fp16")]; + tensor var_1210_cast_fp16 = add(x = var_1206_cast_fp16, y = var_1209_cast_fp16)[name = string("op_1210_cast_fp16")]; + tensor var_1212_begin_0 = const()[name = string("op_1212_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1212_end_0 = const()[name = string("op_1212_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_1212_end_mask_0 = const()[name = string("op_1212_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1212_squeeze_mask_0 = const()[name = string("op_1212_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1212_cast_fp16 = slice_by_index(begin = var_1212_begin_0, end = var_1212_end_0, end_mask = var_1212_end_mask_0, squeeze_mask = var_1212_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1212_cast_fp16")]; + tensor var_1213_cast_fp16 = mul(x = state_37_cast_fp16, y = var_1212_cast_fp16)[name = string("op_1213_cast_fp16")]; + tensor var_1215_begin_0 = const()[name = string("op_1215_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1215_end_0 = const()[name = string("op_1215_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1215_end_mask_0 = const()[name = string("op_1215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1215_squeeze_mask_0 = const()[name = string("op_1215_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1215_cast_fp16 = slice_by_index(begin = var_1215_begin_0, end = var_1215_end_0, end_mask = var_1215_end_mask_0, squeeze_mask = var_1215_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1215_cast_fp16")]; + bool var_1217_transpose_x_1 = const()[name = string("op_1217_transpose_x_1"), val = bool(true)]; + bool var_1217_transpose_y_1 = const()[name = string("op_1217_transpose_y_1"), val = bool(false)]; + tensor var_1217_cast_fp16 = matmul(transpose_x = var_1217_transpose_x_1, transpose_y = var_1217_transpose_y_1, x = var_1215_cast_fp16, y = v_new_37_cast_fp16)[name = string("op_1217_cast_fp16")]; + tensor state_39_cast_fp16 = add(x = var_1213_cast_fp16, y = var_1217_cast_fp16)[name = string("state_39_cast_fp16")]; + tensor var_1220_begin_0 = const()[name = string("op_1220_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1220_end_0 = const()[name = string("op_1220_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1220_end_mask_0 = const()[name = string("op_1220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1220_squeeze_mask_0 = const()[name = string("op_1220_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1220_cast_fp16 = slice_by_index(begin = var_1220_begin_0, end = var_1220_end_0, end_mask = var_1220_end_mask_0, squeeze_mask = var_1220_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1220_cast_fp16")]; + tensor var_1222_begin_0 = const()[name = string("op_1222_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1222_end_0 = const()[name = string("op_1222_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1222_end_mask_0 = const()[name = string("op_1222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1222_squeeze_mask_0 = const()[name = string("op_1222_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1222_cast_fp16 = slice_by_index(begin = var_1222_begin_0, end = var_1222_end_0, end_mask = var_1222_end_mask_0, squeeze_mask = var_1222_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1222_cast_fp16")]; + bool var_1223_transpose_x_0 = const()[name = string("op_1223_transpose_x_0"), val = bool(false)]; + bool var_1223_transpose_y_0 = const()[name = string("op_1223_transpose_y_0"), val = bool(false)]; + tensor var_1223_cast_fp16 = matmul(transpose_x = var_1223_transpose_x_0, transpose_y = var_1223_transpose_y_0, x = var_1222_cast_fp16, y = state_39_cast_fp16)[name = string("op_1223_cast_fp16")]; + tensor v_new_39_cast_fp16 = sub(x = var_1220_cast_fp16, y = var_1223_cast_fp16)[name = string("v_new_39_cast_fp16")]; + tensor var_1226_begin_0 = const()[name = string("op_1226_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1226_end_0 = const()[name = string("op_1226_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1226_end_mask_0 = const()[name = string("op_1226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1226_squeeze_mask_0 = const()[name = string("op_1226_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1226_cast_fp16 = slice_by_index(begin = var_1226_begin_0, end = var_1226_end_0, end_mask = var_1226_end_mask_0, squeeze_mask = var_1226_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1226_cast_fp16")]; + bool var_1227_transpose_x_0 = const()[name = string("op_1227_transpose_x_0"), val = bool(false)]; + bool var_1227_transpose_y_0 = const()[name = string("op_1227_transpose_y_0"), val = bool(false)]; + tensor var_1227_cast_fp16 = matmul(transpose_x = var_1227_transpose_x_0, transpose_y = var_1227_transpose_y_0, x = var_1226_cast_fp16, y = state_39_cast_fp16)[name = string("op_1227_cast_fp16")]; + tensor var_1229_begin_0 = const()[name = string("op_1229_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1229_end_0 = const()[name = string("op_1229_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_1229_end_mask_0 = const()[name = string("op_1229_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1229_squeeze_mask_0 = const()[name = string("op_1229_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1229_cast_fp16 = slice_by_index(begin = var_1229_begin_0, end = var_1229_end_0, end_mask = var_1229_end_mask_0, squeeze_mask = var_1229_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1229_cast_fp16")]; + bool var_1230_transpose_x_0 = const()[name = string("op_1230_transpose_x_0"), val = bool(false)]; + bool var_1230_transpose_y_0 = const()[name = string("op_1230_transpose_y_0"), val = bool(false)]; + tensor var_1230_cast_fp16 = matmul(transpose_x = var_1230_transpose_x_0, transpose_y = var_1230_transpose_y_0, x = var_1229_cast_fp16, y = v_new_39_cast_fp16)[name = string("op_1230_cast_fp16")]; + tensor var_1231_cast_fp16 = add(x = var_1227_cast_fp16, y = var_1230_cast_fp16)[name = string("op_1231_cast_fp16")]; + tensor var_1233_begin_0 = const()[name = string("op_1233_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1233_end_0 = const()[name = string("op_1233_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_1233_end_mask_0 = const()[name = string("op_1233_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1233_squeeze_mask_0 = const()[name = string("op_1233_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1233_cast_fp16 = slice_by_index(begin = var_1233_begin_0, end = var_1233_end_0, end_mask = var_1233_end_mask_0, squeeze_mask = var_1233_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1233_cast_fp16")]; + tensor var_1234_cast_fp16 = mul(x = state_39_cast_fp16, y = var_1233_cast_fp16)[name = string("op_1234_cast_fp16")]; + tensor var_1236_begin_0 = const()[name = string("op_1236_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1236_end_0 = const()[name = string("op_1236_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1236_end_mask_0 = const()[name = string("op_1236_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1236_squeeze_mask_0 = const()[name = string("op_1236_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1236_cast_fp16 = slice_by_index(begin = var_1236_begin_0, end = var_1236_end_0, end_mask = var_1236_end_mask_0, squeeze_mask = var_1236_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1236_cast_fp16")]; + bool var_1238_transpose_x_1 = const()[name = string("op_1238_transpose_x_1"), val = bool(true)]; + bool var_1238_transpose_y_1 = const()[name = string("op_1238_transpose_y_1"), val = bool(false)]; + tensor var_1238_cast_fp16 = matmul(transpose_x = var_1238_transpose_x_1, transpose_y = var_1238_transpose_y_1, x = var_1236_cast_fp16, y = v_new_39_cast_fp16)[name = string("op_1238_cast_fp16")]; + tensor state_41_cast_fp16 = add(x = var_1234_cast_fp16, y = var_1238_cast_fp16)[name = string("state_41_cast_fp16")]; + tensor var_1241_begin_0 = const()[name = string("op_1241_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1241_end_0 = const()[name = string("op_1241_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1241_end_mask_0 = const()[name = string("op_1241_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1241_squeeze_mask_0 = const()[name = string("op_1241_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1241_cast_fp16 = slice_by_index(begin = var_1241_begin_0, end = var_1241_end_0, end_mask = var_1241_end_mask_0, squeeze_mask = var_1241_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1241_cast_fp16")]; + tensor var_1243_begin_0 = const()[name = string("op_1243_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1243_end_0 = const()[name = string("op_1243_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1243_end_mask_0 = const()[name = string("op_1243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1243_squeeze_mask_0 = const()[name = string("op_1243_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1243_cast_fp16 = slice_by_index(begin = var_1243_begin_0, end = var_1243_end_0, end_mask = var_1243_end_mask_0, squeeze_mask = var_1243_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1243_cast_fp16")]; + bool var_1244_transpose_x_0 = const()[name = string("op_1244_transpose_x_0"), val = bool(false)]; + bool var_1244_transpose_y_0 = const()[name = string("op_1244_transpose_y_0"), val = bool(false)]; + tensor var_1244_cast_fp16 = matmul(transpose_x = var_1244_transpose_x_0, transpose_y = var_1244_transpose_y_0, x = var_1243_cast_fp16, y = state_41_cast_fp16)[name = string("op_1244_cast_fp16")]; + tensor v_new_41_cast_fp16 = sub(x = var_1241_cast_fp16, y = var_1244_cast_fp16)[name = string("v_new_41_cast_fp16")]; + tensor var_1247_begin_0 = const()[name = string("op_1247_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1247_end_0 = const()[name = string("op_1247_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1247_end_mask_0 = const()[name = string("op_1247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1247_squeeze_mask_0 = const()[name = string("op_1247_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1247_cast_fp16 = slice_by_index(begin = var_1247_begin_0, end = var_1247_end_0, end_mask = var_1247_end_mask_0, squeeze_mask = var_1247_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1247_cast_fp16")]; + bool var_1248_transpose_x_0 = const()[name = string("op_1248_transpose_x_0"), val = bool(false)]; + bool var_1248_transpose_y_0 = const()[name = string("op_1248_transpose_y_0"), val = bool(false)]; + tensor var_1248_cast_fp16 = matmul(transpose_x = var_1248_transpose_x_0, transpose_y = var_1248_transpose_y_0, x = var_1247_cast_fp16, y = state_41_cast_fp16)[name = string("op_1248_cast_fp16")]; + tensor var_1250_begin_0 = const()[name = string("op_1250_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1250_end_0 = const()[name = string("op_1250_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_1250_end_mask_0 = const()[name = string("op_1250_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1250_squeeze_mask_0 = const()[name = string("op_1250_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1250_cast_fp16 = slice_by_index(begin = var_1250_begin_0, end = var_1250_end_0, end_mask = var_1250_end_mask_0, squeeze_mask = var_1250_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1250_cast_fp16")]; + bool var_1251_transpose_x_0 = const()[name = string("op_1251_transpose_x_0"), val = bool(false)]; + bool var_1251_transpose_y_0 = const()[name = string("op_1251_transpose_y_0"), val = bool(false)]; + tensor var_1251_cast_fp16 = matmul(transpose_x = var_1251_transpose_x_0, transpose_y = var_1251_transpose_y_0, x = var_1250_cast_fp16, y = v_new_41_cast_fp16)[name = string("op_1251_cast_fp16")]; + tensor var_1252_cast_fp16 = add(x = var_1248_cast_fp16, y = var_1251_cast_fp16)[name = string("op_1252_cast_fp16")]; + tensor var_1254_begin_0 = const()[name = string("op_1254_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1254_end_0 = const()[name = string("op_1254_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_1254_end_mask_0 = const()[name = string("op_1254_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1254_squeeze_mask_0 = const()[name = string("op_1254_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1254_cast_fp16 = slice_by_index(begin = var_1254_begin_0, end = var_1254_end_0, end_mask = var_1254_end_mask_0, squeeze_mask = var_1254_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1254_cast_fp16")]; + tensor var_1255_cast_fp16 = mul(x = state_41_cast_fp16, y = var_1254_cast_fp16)[name = string("op_1255_cast_fp16")]; + tensor var_1257_begin_0 = const()[name = string("op_1257_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1257_end_0 = const()[name = string("op_1257_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1257_end_mask_0 = const()[name = string("op_1257_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1257_squeeze_mask_0 = const()[name = string("op_1257_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1257_cast_fp16 = slice_by_index(begin = var_1257_begin_0, end = var_1257_end_0, end_mask = var_1257_end_mask_0, squeeze_mask = var_1257_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1257_cast_fp16")]; + bool var_1259_transpose_x_1 = const()[name = string("op_1259_transpose_x_1"), val = bool(true)]; + bool var_1259_transpose_y_1 = const()[name = string("op_1259_transpose_y_1"), val = bool(false)]; + tensor var_1259_cast_fp16 = matmul(transpose_x = var_1259_transpose_x_1, transpose_y = var_1259_transpose_y_1, x = var_1257_cast_fp16, y = v_new_41_cast_fp16)[name = string("op_1259_cast_fp16")]; + tensor state_43_cast_fp16 = add(x = var_1255_cast_fp16, y = var_1259_cast_fp16)[name = string("state_43_cast_fp16")]; + tensor var_1262_begin_0 = const()[name = string("op_1262_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1262_end_0 = const()[name = string("op_1262_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_1262_end_mask_0 = const()[name = string("op_1262_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1262_squeeze_mask_0 = const()[name = string("op_1262_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, squeeze_mask = var_1262_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1262_cast_fp16")]; + tensor var_1264_begin_0 = const()[name = string("op_1264_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1264_end_0 = const()[name = string("op_1264_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_1264_end_mask_0 = const()[name = string("op_1264_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1264_squeeze_mask_0 = const()[name = string("op_1264_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1264_cast_fp16 = slice_by_index(begin = var_1264_begin_0, end = var_1264_end_0, end_mask = var_1264_end_mask_0, squeeze_mask = var_1264_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1264_cast_fp16")]; + bool var_1265_transpose_x_0 = const()[name = string("op_1265_transpose_x_0"), val = bool(false)]; + bool var_1265_transpose_y_0 = const()[name = string("op_1265_transpose_y_0"), val = bool(false)]; + tensor var_1265_cast_fp16 = matmul(transpose_x = var_1265_transpose_x_0, transpose_y = var_1265_transpose_y_0, x = var_1264_cast_fp16, y = state_43_cast_fp16)[name = string("op_1265_cast_fp16")]; + tensor v_new_43_cast_fp16 = sub(x = var_1262_cast_fp16, y = var_1265_cast_fp16)[name = string("v_new_43_cast_fp16")]; + tensor var_1268_begin_0 = const()[name = string("op_1268_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1268_end_0 = const()[name = string("op_1268_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_1268_end_mask_0 = const()[name = string("op_1268_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1268_squeeze_mask_0 = const()[name = string("op_1268_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1268_cast_fp16 = slice_by_index(begin = var_1268_begin_0, end = var_1268_end_0, end_mask = var_1268_end_mask_0, squeeze_mask = var_1268_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1268_cast_fp16")]; + bool var_1269_transpose_x_0 = const()[name = string("op_1269_transpose_x_0"), val = bool(false)]; + bool var_1269_transpose_y_0 = const()[name = string("op_1269_transpose_y_0"), val = bool(false)]; + tensor var_1269_cast_fp16 = matmul(transpose_x = var_1269_transpose_x_0, transpose_y = var_1269_transpose_y_0, x = var_1268_cast_fp16, y = state_43_cast_fp16)[name = string("op_1269_cast_fp16")]; + tensor var_1271_begin_0 = const()[name = string("op_1271_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1271_end_0 = const()[name = string("op_1271_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_1271_end_mask_0 = const()[name = string("op_1271_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1271_squeeze_mask_0 = const()[name = string("op_1271_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1271_cast_fp16 = slice_by_index(begin = var_1271_begin_0, end = var_1271_end_0, end_mask = var_1271_end_mask_0, squeeze_mask = var_1271_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1271_cast_fp16")]; + bool var_1272_transpose_x_0 = const()[name = string("op_1272_transpose_x_0"), val = bool(false)]; + bool var_1272_transpose_y_0 = const()[name = string("op_1272_transpose_y_0"), val = bool(false)]; + tensor var_1272_cast_fp16 = matmul(transpose_x = var_1272_transpose_x_0, transpose_y = var_1272_transpose_y_0, x = var_1271_cast_fp16, y = v_new_43_cast_fp16)[name = string("op_1272_cast_fp16")]; + tensor var_1273_cast_fp16 = add(x = var_1269_cast_fp16, y = var_1272_cast_fp16)[name = string("op_1273_cast_fp16")]; + tensor var_1275_begin_0 = const()[name = string("op_1275_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1275_end_0 = const()[name = string("op_1275_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_1275_end_mask_0 = const()[name = string("op_1275_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1275_squeeze_mask_0 = const()[name = string("op_1275_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1275_cast_fp16 = slice_by_index(begin = var_1275_begin_0, end = var_1275_end_0, end_mask = var_1275_end_mask_0, squeeze_mask = var_1275_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1275_cast_fp16")]; + tensor var_1276_cast_fp16 = mul(x = state_43_cast_fp16, y = var_1275_cast_fp16)[name = string("op_1276_cast_fp16")]; + tensor var_1278_begin_0 = const()[name = string("op_1278_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_1278_end_0 = const()[name = string("op_1278_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_1278_end_mask_0 = const()[name = string("op_1278_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1278_squeeze_mask_0 = const()[name = string("op_1278_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1278_cast_fp16 = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, squeeze_mask = var_1278_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1278_cast_fp16")]; + bool var_1280_transpose_x_1 = const()[name = string("op_1280_transpose_x_1"), val = bool(true)]; + bool var_1280_transpose_y_1 = const()[name = string("op_1280_transpose_y_1"), val = bool(false)]; + tensor var_1280_cast_fp16 = matmul(transpose_x = var_1280_transpose_x_1, transpose_y = var_1280_transpose_y_1, x = var_1278_cast_fp16, y = v_new_43_cast_fp16)[name = string("op_1280_cast_fp16")]; + tensor state_45_cast_fp16 = add(x = var_1276_cast_fp16, y = var_1280_cast_fp16)[name = string("state_45_cast_fp16")]; + tensor var_1283_begin_0 = const()[name = string("op_1283_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1283_end_0 = const()[name = string("op_1283_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_1283_end_mask_0 = const()[name = string("op_1283_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1283_squeeze_mask_0 = const()[name = string("op_1283_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1283_cast_fp16 = slice_by_index(begin = var_1283_begin_0, end = var_1283_end_0, end_mask = var_1283_end_mask_0, squeeze_mask = var_1283_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1283_cast_fp16")]; + tensor var_1285_begin_0 = const()[name = string("op_1285_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1285_end_0 = const()[name = string("op_1285_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_1285_end_mask_0 = const()[name = string("op_1285_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1285_squeeze_mask_0 = const()[name = string("op_1285_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1285_cast_fp16 = slice_by_index(begin = var_1285_begin_0, end = var_1285_end_0, end_mask = var_1285_end_mask_0, squeeze_mask = var_1285_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1285_cast_fp16")]; + bool var_1286_transpose_x_0 = const()[name = string("op_1286_transpose_x_0"), val = bool(false)]; + bool var_1286_transpose_y_0 = const()[name = string("op_1286_transpose_y_0"), val = bool(false)]; + tensor var_1286_cast_fp16 = matmul(transpose_x = var_1286_transpose_x_0, transpose_y = var_1286_transpose_y_0, x = var_1285_cast_fp16, y = state_45_cast_fp16)[name = string("op_1286_cast_fp16")]; + tensor v_new_45_cast_fp16 = sub(x = var_1283_cast_fp16, y = var_1286_cast_fp16)[name = string("v_new_45_cast_fp16")]; + tensor var_1289_begin_0 = const()[name = string("op_1289_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1289_end_0 = const()[name = string("op_1289_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_1289_end_mask_0 = const()[name = string("op_1289_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1289_squeeze_mask_0 = const()[name = string("op_1289_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1289_cast_fp16 = slice_by_index(begin = var_1289_begin_0, end = var_1289_end_0, end_mask = var_1289_end_mask_0, squeeze_mask = var_1289_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1289_cast_fp16")]; + bool var_1290_transpose_x_0 = const()[name = string("op_1290_transpose_x_0"), val = bool(false)]; + bool var_1290_transpose_y_0 = const()[name = string("op_1290_transpose_y_0"), val = bool(false)]; + tensor var_1290_cast_fp16 = matmul(transpose_x = var_1290_transpose_x_0, transpose_y = var_1290_transpose_y_0, x = var_1289_cast_fp16, y = state_45_cast_fp16)[name = string("op_1290_cast_fp16")]; + tensor var_1292_begin_0 = const()[name = string("op_1292_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1292_end_0 = const()[name = string("op_1292_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_1292_end_mask_0 = const()[name = string("op_1292_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1292_squeeze_mask_0 = const()[name = string("op_1292_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1292_cast_fp16 = slice_by_index(begin = var_1292_begin_0, end = var_1292_end_0, end_mask = var_1292_end_mask_0, squeeze_mask = var_1292_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1292_cast_fp16")]; + bool var_1293_transpose_x_0 = const()[name = string("op_1293_transpose_x_0"), val = bool(false)]; + bool var_1293_transpose_y_0 = const()[name = string("op_1293_transpose_y_0"), val = bool(false)]; + tensor var_1293_cast_fp16 = matmul(transpose_x = var_1293_transpose_x_0, transpose_y = var_1293_transpose_y_0, x = var_1292_cast_fp16, y = v_new_45_cast_fp16)[name = string("op_1293_cast_fp16")]; + tensor var_1294_cast_fp16 = add(x = var_1290_cast_fp16, y = var_1293_cast_fp16)[name = string("op_1294_cast_fp16")]; + tensor var_1296_begin_0 = const()[name = string("op_1296_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1296_end_0 = const()[name = string("op_1296_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_1296_end_mask_0 = const()[name = string("op_1296_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1296_squeeze_mask_0 = const()[name = string("op_1296_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1296_cast_fp16 = slice_by_index(begin = var_1296_begin_0, end = var_1296_end_0, end_mask = var_1296_end_mask_0, squeeze_mask = var_1296_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1296_cast_fp16")]; + tensor var_1297_cast_fp16 = mul(x = state_45_cast_fp16, y = var_1296_cast_fp16)[name = string("op_1297_cast_fp16")]; + tensor var_1299_begin_0 = const()[name = string("op_1299_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_1299_end_0 = const()[name = string("op_1299_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_1299_end_mask_0 = const()[name = string("op_1299_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1299_squeeze_mask_0 = const()[name = string("op_1299_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1299_cast_fp16 = slice_by_index(begin = var_1299_begin_0, end = var_1299_end_0, end_mask = var_1299_end_mask_0, squeeze_mask = var_1299_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1299_cast_fp16")]; + bool var_1301_transpose_x_1 = const()[name = string("op_1301_transpose_x_1"), val = bool(true)]; + bool var_1301_transpose_y_1 = const()[name = string("op_1301_transpose_y_1"), val = bool(false)]; + tensor var_1301_cast_fp16 = matmul(transpose_x = var_1301_transpose_x_1, transpose_y = var_1301_transpose_y_1, x = var_1299_cast_fp16, y = v_new_45_cast_fp16)[name = string("op_1301_cast_fp16")]; + tensor state_47_cast_fp16 = add(x = var_1297_cast_fp16, y = var_1301_cast_fp16)[name = string("state_47_cast_fp16")]; + tensor var_1304_begin_0 = const()[name = string("op_1304_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1304_end_0 = const()[name = string("op_1304_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_1304_end_mask_0 = const()[name = string("op_1304_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1304_squeeze_mask_0 = const()[name = string("op_1304_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1304_cast_fp16 = slice_by_index(begin = var_1304_begin_0, end = var_1304_end_0, end_mask = var_1304_end_mask_0, squeeze_mask = var_1304_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1304_cast_fp16")]; + tensor var_1306_begin_0 = const()[name = string("op_1306_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1306_end_0 = const()[name = string("op_1306_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_1306_end_mask_0 = const()[name = string("op_1306_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1306_squeeze_mask_0 = const()[name = string("op_1306_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1306_cast_fp16 = slice_by_index(begin = var_1306_begin_0, end = var_1306_end_0, end_mask = var_1306_end_mask_0, squeeze_mask = var_1306_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1306_cast_fp16")]; + bool var_1307_transpose_x_0 = const()[name = string("op_1307_transpose_x_0"), val = bool(false)]; + bool var_1307_transpose_y_0 = const()[name = string("op_1307_transpose_y_0"), val = bool(false)]; + tensor var_1307_cast_fp16 = matmul(transpose_x = var_1307_transpose_x_0, transpose_y = var_1307_transpose_y_0, x = var_1306_cast_fp16, y = state_47_cast_fp16)[name = string("op_1307_cast_fp16")]; + tensor v_new_47_cast_fp16 = sub(x = var_1304_cast_fp16, y = var_1307_cast_fp16)[name = string("v_new_47_cast_fp16")]; + tensor var_1310_begin_0 = const()[name = string("op_1310_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1310_end_0 = const()[name = string("op_1310_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_1310_end_mask_0 = const()[name = string("op_1310_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1310_squeeze_mask_0 = const()[name = string("op_1310_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1310_cast_fp16 = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, squeeze_mask = var_1310_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1310_cast_fp16")]; + bool var_1311_transpose_x_0 = const()[name = string("op_1311_transpose_x_0"), val = bool(false)]; + bool var_1311_transpose_y_0 = const()[name = string("op_1311_transpose_y_0"), val = bool(false)]; + tensor var_1311_cast_fp16 = matmul(transpose_x = var_1311_transpose_x_0, transpose_y = var_1311_transpose_y_0, x = var_1310_cast_fp16, y = state_47_cast_fp16)[name = string("op_1311_cast_fp16")]; + tensor var_1313_begin_0 = const()[name = string("op_1313_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1313_end_0 = const()[name = string("op_1313_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_1313_end_mask_0 = const()[name = string("op_1313_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1313_squeeze_mask_0 = const()[name = string("op_1313_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1313_cast_fp16 = slice_by_index(begin = var_1313_begin_0, end = var_1313_end_0, end_mask = var_1313_end_mask_0, squeeze_mask = var_1313_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1313_cast_fp16")]; + bool var_1314_transpose_x_0 = const()[name = string("op_1314_transpose_x_0"), val = bool(false)]; + bool var_1314_transpose_y_0 = const()[name = string("op_1314_transpose_y_0"), val = bool(false)]; + tensor var_1314_cast_fp16 = matmul(transpose_x = var_1314_transpose_x_0, transpose_y = var_1314_transpose_y_0, x = var_1313_cast_fp16, y = v_new_47_cast_fp16)[name = string("op_1314_cast_fp16")]; + tensor var_1315_cast_fp16 = add(x = var_1311_cast_fp16, y = var_1314_cast_fp16)[name = string("op_1315_cast_fp16")]; + tensor var_1317_begin_0 = const()[name = string("op_1317_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1317_end_0 = const()[name = string("op_1317_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_1317_end_mask_0 = const()[name = string("op_1317_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1317_squeeze_mask_0 = const()[name = string("op_1317_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1317_cast_fp16 = slice_by_index(begin = var_1317_begin_0, end = var_1317_end_0, end_mask = var_1317_end_mask_0, squeeze_mask = var_1317_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1317_cast_fp16")]; + tensor var_1318_cast_fp16 = mul(x = state_47_cast_fp16, y = var_1317_cast_fp16)[name = string("op_1318_cast_fp16")]; + tensor var_1320_begin_0 = const()[name = string("op_1320_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_1320_end_0 = const()[name = string("op_1320_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_1320_end_mask_0 = const()[name = string("op_1320_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1320_squeeze_mask_0 = const()[name = string("op_1320_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1320_cast_fp16 = slice_by_index(begin = var_1320_begin_0, end = var_1320_end_0, end_mask = var_1320_end_mask_0, squeeze_mask = var_1320_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1320_cast_fp16")]; + bool var_1322_transpose_x_1 = const()[name = string("op_1322_transpose_x_1"), val = bool(true)]; + bool var_1322_transpose_y_1 = const()[name = string("op_1322_transpose_y_1"), val = bool(false)]; + tensor var_1322_cast_fp16 = matmul(transpose_x = var_1322_transpose_x_1, transpose_y = var_1322_transpose_y_1, x = var_1320_cast_fp16, y = v_new_47_cast_fp16)[name = string("op_1322_cast_fp16")]; + tensor state_49_cast_fp16 = add(x = var_1318_cast_fp16, y = var_1322_cast_fp16)[name = string("state_49_cast_fp16")]; + tensor var_1325_begin_0 = const()[name = string("op_1325_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1325_end_0 = const()[name = string("op_1325_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_1325_end_mask_0 = const()[name = string("op_1325_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1325_squeeze_mask_0 = const()[name = string("op_1325_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1325_cast_fp16 = slice_by_index(begin = var_1325_begin_0, end = var_1325_end_0, end_mask = var_1325_end_mask_0, squeeze_mask = var_1325_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1325_cast_fp16")]; + tensor var_1327_begin_0 = const()[name = string("op_1327_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1327_end_0 = const()[name = string("op_1327_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_1327_end_mask_0 = const()[name = string("op_1327_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1327_squeeze_mask_0 = const()[name = string("op_1327_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1327_cast_fp16 = slice_by_index(begin = var_1327_begin_0, end = var_1327_end_0, end_mask = var_1327_end_mask_0, squeeze_mask = var_1327_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1327_cast_fp16")]; + bool var_1328_transpose_x_0 = const()[name = string("op_1328_transpose_x_0"), val = bool(false)]; + bool var_1328_transpose_y_0 = const()[name = string("op_1328_transpose_y_0"), val = bool(false)]; + tensor var_1328_cast_fp16 = matmul(transpose_x = var_1328_transpose_x_0, transpose_y = var_1328_transpose_y_0, x = var_1327_cast_fp16, y = state_49_cast_fp16)[name = string("op_1328_cast_fp16")]; + tensor v_new_49_cast_fp16 = sub(x = var_1325_cast_fp16, y = var_1328_cast_fp16)[name = string("v_new_49_cast_fp16")]; + tensor var_1331_begin_0 = const()[name = string("op_1331_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1331_end_0 = const()[name = string("op_1331_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_1331_end_mask_0 = const()[name = string("op_1331_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1331_squeeze_mask_0 = const()[name = string("op_1331_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1331_cast_fp16 = slice_by_index(begin = var_1331_begin_0, end = var_1331_end_0, end_mask = var_1331_end_mask_0, squeeze_mask = var_1331_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1331_cast_fp16")]; + bool var_1332_transpose_x_0 = const()[name = string("op_1332_transpose_x_0"), val = bool(false)]; + bool var_1332_transpose_y_0 = const()[name = string("op_1332_transpose_y_0"), val = bool(false)]; + tensor var_1332_cast_fp16 = matmul(transpose_x = var_1332_transpose_x_0, transpose_y = var_1332_transpose_y_0, x = var_1331_cast_fp16, y = state_49_cast_fp16)[name = string("op_1332_cast_fp16")]; + tensor var_1334_begin_0 = const()[name = string("op_1334_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1334_end_0 = const()[name = string("op_1334_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_1334_end_mask_0 = const()[name = string("op_1334_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1334_squeeze_mask_0 = const()[name = string("op_1334_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1334_cast_fp16 = slice_by_index(begin = var_1334_begin_0, end = var_1334_end_0, end_mask = var_1334_end_mask_0, squeeze_mask = var_1334_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1334_cast_fp16")]; + bool var_1335_transpose_x_0 = const()[name = string("op_1335_transpose_x_0"), val = bool(false)]; + bool var_1335_transpose_y_0 = const()[name = string("op_1335_transpose_y_0"), val = bool(false)]; + tensor var_1335_cast_fp16 = matmul(transpose_x = var_1335_transpose_x_0, transpose_y = var_1335_transpose_y_0, x = var_1334_cast_fp16, y = v_new_49_cast_fp16)[name = string("op_1335_cast_fp16")]; + tensor var_1336_cast_fp16 = add(x = var_1332_cast_fp16, y = var_1335_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1338_begin_0 = const()[name = string("op_1338_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1338_end_0 = const()[name = string("op_1338_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_1338_end_mask_0 = const()[name = string("op_1338_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1338_squeeze_mask_0 = const()[name = string("op_1338_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1338_cast_fp16 = slice_by_index(begin = var_1338_begin_0, end = var_1338_end_0, end_mask = var_1338_end_mask_0, squeeze_mask = var_1338_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1338_cast_fp16")]; + tensor var_1339_cast_fp16 = mul(x = state_49_cast_fp16, y = var_1338_cast_fp16)[name = string("op_1339_cast_fp16")]; + tensor var_1341_begin_0 = const()[name = string("op_1341_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_1341_end_0 = const()[name = string("op_1341_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_1341_end_mask_0 = const()[name = string("op_1341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1341_squeeze_mask_0 = const()[name = string("op_1341_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1341_cast_fp16 = slice_by_index(begin = var_1341_begin_0, end = var_1341_end_0, end_mask = var_1341_end_mask_0, squeeze_mask = var_1341_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1341_cast_fp16")]; + bool var_1343_transpose_x_1 = const()[name = string("op_1343_transpose_x_1"), val = bool(true)]; + bool var_1343_transpose_y_1 = const()[name = string("op_1343_transpose_y_1"), val = bool(false)]; + tensor var_1343_cast_fp16 = matmul(transpose_x = var_1343_transpose_x_1, transpose_y = var_1343_transpose_y_1, x = var_1341_cast_fp16, y = v_new_49_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor state_51_cast_fp16 = add(x = var_1339_cast_fp16, y = var_1343_cast_fp16)[name = string("state_51_cast_fp16")]; + tensor var_1346_begin_0 = const()[name = string("op_1346_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1346_end_0 = const()[name = string("op_1346_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_1346_end_mask_0 = const()[name = string("op_1346_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1346_squeeze_mask_0 = const()[name = string("op_1346_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1346_cast_fp16 = slice_by_index(begin = var_1346_begin_0, end = var_1346_end_0, end_mask = var_1346_end_mask_0, squeeze_mask = var_1346_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1346_cast_fp16")]; + tensor var_1348_begin_0 = const()[name = string("op_1348_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1348_end_0 = const()[name = string("op_1348_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_1348_end_mask_0 = const()[name = string("op_1348_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1348_squeeze_mask_0 = const()[name = string("op_1348_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1348_cast_fp16 = slice_by_index(begin = var_1348_begin_0, end = var_1348_end_0, end_mask = var_1348_end_mask_0, squeeze_mask = var_1348_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1348_cast_fp16")]; + bool var_1349_transpose_x_0 = const()[name = string("op_1349_transpose_x_0"), val = bool(false)]; + bool var_1349_transpose_y_0 = const()[name = string("op_1349_transpose_y_0"), val = bool(false)]; + tensor var_1349_cast_fp16 = matmul(transpose_x = var_1349_transpose_x_0, transpose_y = var_1349_transpose_y_0, x = var_1348_cast_fp16, y = state_51_cast_fp16)[name = string("op_1349_cast_fp16")]; + tensor v_new_51_cast_fp16 = sub(x = var_1346_cast_fp16, y = var_1349_cast_fp16)[name = string("v_new_51_cast_fp16")]; + tensor var_1352_begin_0 = const()[name = string("op_1352_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1352_end_0 = const()[name = string("op_1352_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_1352_end_mask_0 = const()[name = string("op_1352_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1352_squeeze_mask_0 = const()[name = string("op_1352_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1352_cast_fp16 = slice_by_index(begin = var_1352_begin_0, end = var_1352_end_0, end_mask = var_1352_end_mask_0, squeeze_mask = var_1352_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1352_cast_fp16")]; + bool var_1353_transpose_x_0 = const()[name = string("op_1353_transpose_x_0"), val = bool(false)]; + bool var_1353_transpose_y_0 = const()[name = string("op_1353_transpose_y_0"), val = bool(false)]; + tensor var_1353_cast_fp16 = matmul(transpose_x = var_1353_transpose_x_0, transpose_y = var_1353_transpose_y_0, x = var_1352_cast_fp16, y = state_51_cast_fp16)[name = string("op_1353_cast_fp16")]; + tensor var_1355_begin_0 = const()[name = string("op_1355_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1355_end_0 = const()[name = string("op_1355_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_1355_end_mask_0 = const()[name = string("op_1355_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1355_squeeze_mask_0 = const()[name = string("op_1355_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1355_cast_fp16 = slice_by_index(begin = var_1355_begin_0, end = var_1355_end_0, end_mask = var_1355_end_mask_0, squeeze_mask = var_1355_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1355_cast_fp16")]; + bool var_1356_transpose_x_0 = const()[name = string("op_1356_transpose_x_0"), val = bool(false)]; + bool var_1356_transpose_y_0 = const()[name = string("op_1356_transpose_y_0"), val = bool(false)]; + tensor var_1356_cast_fp16 = matmul(transpose_x = var_1356_transpose_x_0, transpose_y = var_1356_transpose_y_0, x = var_1355_cast_fp16, y = v_new_51_cast_fp16)[name = string("op_1356_cast_fp16")]; + tensor var_1357_cast_fp16 = add(x = var_1353_cast_fp16, y = var_1356_cast_fp16)[name = string("op_1357_cast_fp16")]; + tensor var_1359_begin_0 = const()[name = string("op_1359_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1359_end_0 = const()[name = string("op_1359_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_1359_end_mask_0 = const()[name = string("op_1359_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1359_squeeze_mask_0 = const()[name = string("op_1359_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1359_cast_fp16 = slice_by_index(begin = var_1359_begin_0, end = var_1359_end_0, end_mask = var_1359_end_mask_0, squeeze_mask = var_1359_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1359_cast_fp16")]; + tensor var_1360_cast_fp16 = mul(x = state_51_cast_fp16, y = var_1359_cast_fp16)[name = string("op_1360_cast_fp16")]; + tensor var_1362_begin_0 = const()[name = string("op_1362_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_1362_end_0 = const()[name = string("op_1362_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_1362_end_mask_0 = const()[name = string("op_1362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1362_squeeze_mask_0 = const()[name = string("op_1362_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1362_cast_fp16 = slice_by_index(begin = var_1362_begin_0, end = var_1362_end_0, end_mask = var_1362_end_mask_0, squeeze_mask = var_1362_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1362_cast_fp16")]; + bool var_1364_transpose_x_1 = const()[name = string("op_1364_transpose_x_1"), val = bool(true)]; + bool var_1364_transpose_y_1 = const()[name = string("op_1364_transpose_y_1"), val = bool(false)]; + tensor var_1364_cast_fp16 = matmul(transpose_x = var_1364_transpose_x_1, transpose_y = var_1364_transpose_y_1, x = var_1362_cast_fp16, y = v_new_51_cast_fp16)[name = string("op_1364_cast_fp16")]; + tensor state_53_cast_fp16 = add(x = var_1360_cast_fp16, y = var_1364_cast_fp16)[name = string("state_53_cast_fp16")]; + tensor var_1367_begin_0 = const()[name = string("op_1367_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1367_end_0 = const()[name = string("op_1367_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_1367_end_mask_0 = const()[name = string("op_1367_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1367_squeeze_mask_0 = const()[name = string("op_1367_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1367_cast_fp16 = slice_by_index(begin = var_1367_begin_0, end = var_1367_end_0, end_mask = var_1367_end_mask_0, squeeze_mask = var_1367_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1367_cast_fp16")]; + tensor var_1369_begin_0 = const()[name = string("op_1369_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1369_end_0 = const()[name = string("op_1369_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_1369_end_mask_0 = const()[name = string("op_1369_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1369_squeeze_mask_0 = const()[name = string("op_1369_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1369_cast_fp16 = slice_by_index(begin = var_1369_begin_0, end = var_1369_end_0, end_mask = var_1369_end_mask_0, squeeze_mask = var_1369_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1369_cast_fp16")]; + bool var_1370_transpose_x_0 = const()[name = string("op_1370_transpose_x_0"), val = bool(false)]; + bool var_1370_transpose_y_0 = const()[name = string("op_1370_transpose_y_0"), val = bool(false)]; + tensor var_1370_cast_fp16 = matmul(transpose_x = var_1370_transpose_x_0, transpose_y = var_1370_transpose_y_0, x = var_1369_cast_fp16, y = state_53_cast_fp16)[name = string("op_1370_cast_fp16")]; + tensor v_new_53_cast_fp16 = sub(x = var_1367_cast_fp16, y = var_1370_cast_fp16)[name = string("v_new_53_cast_fp16")]; + tensor var_1373_begin_0 = const()[name = string("op_1373_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1373_end_0 = const()[name = string("op_1373_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_1373_end_mask_0 = const()[name = string("op_1373_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1373_squeeze_mask_0 = const()[name = string("op_1373_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1373_cast_fp16 = slice_by_index(begin = var_1373_begin_0, end = var_1373_end_0, end_mask = var_1373_end_mask_0, squeeze_mask = var_1373_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1373_cast_fp16")]; + bool var_1374_transpose_x_0 = const()[name = string("op_1374_transpose_x_0"), val = bool(false)]; + bool var_1374_transpose_y_0 = const()[name = string("op_1374_transpose_y_0"), val = bool(false)]; + tensor var_1374_cast_fp16 = matmul(transpose_x = var_1374_transpose_x_0, transpose_y = var_1374_transpose_y_0, x = var_1373_cast_fp16, y = state_53_cast_fp16)[name = string("op_1374_cast_fp16")]; + tensor var_1376_begin_0 = const()[name = string("op_1376_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1376_end_0 = const()[name = string("op_1376_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_1376_end_mask_0 = const()[name = string("op_1376_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1376_squeeze_mask_0 = const()[name = string("op_1376_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1376_cast_fp16 = slice_by_index(begin = var_1376_begin_0, end = var_1376_end_0, end_mask = var_1376_end_mask_0, squeeze_mask = var_1376_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1376_cast_fp16")]; + bool var_1377_transpose_x_0 = const()[name = string("op_1377_transpose_x_0"), val = bool(false)]; + bool var_1377_transpose_y_0 = const()[name = string("op_1377_transpose_y_0"), val = bool(false)]; + tensor var_1377_cast_fp16 = matmul(transpose_x = var_1377_transpose_x_0, transpose_y = var_1377_transpose_y_0, x = var_1376_cast_fp16, y = v_new_53_cast_fp16)[name = string("op_1377_cast_fp16")]; + tensor var_1378_cast_fp16 = add(x = var_1374_cast_fp16, y = var_1377_cast_fp16)[name = string("op_1378_cast_fp16")]; + tensor var_1380_begin_0 = const()[name = string("op_1380_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1380_end_0 = const()[name = string("op_1380_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_1380_end_mask_0 = const()[name = string("op_1380_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1380_squeeze_mask_0 = const()[name = string("op_1380_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1380_cast_fp16 = slice_by_index(begin = var_1380_begin_0, end = var_1380_end_0, end_mask = var_1380_end_mask_0, squeeze_mask = var_1380_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1380_cast_fp16")]; + tensor var_1381_cast_fp16 = mul(x = state_53_cast_fp16, y = var_1380_cast_fp16)[name = string("op_1381_cast_fp16")]; + tensor var_1383_begin_0 = const()[name = string("op_1383_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_1383_end_0 = const()[name = string("op_1383_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_1383_end_mask_0 = const()[name = string("op_1383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1383_squeeze_mask_0 = const()[name = string("op_1383_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1383_cast_fp16 = slice_by_index(begin = var_1383_begin_0, end = var_1383_end_0, end_mask = var_1383_end_mask_0, squeeze_mask = var_1383_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1383_cast_fp16")]; + bool var_1385_transpose_x_1 = const()[name = string("op_1385_transpose_x_1"), val = bool(true)]; + bool var_1385_transpose_y_1 = const()[name = string("op_1385_transpose_y_1"), val = bool(false)]; + tensor var_1385_cast_fp16 = matmul(transpose_x = var_1385_transpose_x_1, transpose_y = var_1385_transpose_y_1, x = var_1383_cast_fp16, y = v_new_53_cast_fp16)[name = string("op_1385_cast_fp16")]; + tensor state_55_cast_fp16 = add(x = var_1381_cast_fp16, y = var_1385_cast_fp16)[name = string("state_55_cast_fp16")]; + tensor var_1388_begin_0 = const()[name = string("op_1388_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1388_end_0 = const()[name = string("op_1388_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_1388_end_mask_0 = const()[name = string("op_1388_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1388_squeeze_mask_0 = const()[name = string("op_1388_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1388_cast_fp16 = slice_by_index(begin = var_1388_begin_0, end = var_1388_end_0, end_mask = var_1388_end_mask_0, squeeze_mask = var_1388_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1388_cast_fp16")]; + tensor var_1390_begin_0 = const()[name = string("op_1390_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1390_end_0 = const()[name = string("op_1390_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_1390_end_mask_0 = const()[name = string("op_1390_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1390_squeeze_mask_0 = const()[name = string("op_1390_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1390_cast_fp16 = slice_by_index(begin = var_1390_begin_0, end = var_1390_end_0, end_mask = var_1390_end_mask_0, squeeze_mask = var_1390_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1390_cast_fp16")]; + bool var_1391_transpose_x_0 = const()[name = string("op_1391_transpose_x_0"), val = bool(false)]; + bool var_1391_transpose_y_0 = const()[name = string("op_1391_transpose_y_0"), val = bool(false)]; + tensor var_1391_cast_fp16 = matmul(transpose_x = var_1391_transpose_x_0, transpose_y = var_1391_transpose_y_0, x = var_1390_cast_fp16, y = state_55_cast_fp16)[name = string("op_1391_cast_fp16")]; + tensor v_new_55_cast_fp16 = sub(x = var_1388_cast_fp16, y = var_1391_cast_fp16)[name = string("v_new_55_cast_fp16")]; + tensor var_1394_begin_0 = const()[name = string("op_1394_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1394_end_0 = const()[name = string("op_1394_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_1394_end_mask_0 = const()[name = string("op_1394_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1394_squeeze_mask_0 = const()[name = string("op_1394_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1394_cast_fp16 = slice_by_index(begin = var_1394_begin_0, end = var_1394_end_0, end_mask = var_1394_end_mask_0, squeeze_mask = var_1394_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1394_cast_fp16")]; + bool var_1395_transpose_x_0 = const()[name = string("op_1395_transpose_x_0"), val = bool(false)]; + bool var_1395_transpose_y_0 = const()[name = string("op_1395_transpose_y_0"), val = bool(false)]; + tensor var_1395_cast_fp16 = matmul(transpose_x = var_1395_transpose_x_0, transpose_y = var_1395_transpose_y_0, x = var_1394_cast_fp16, y = state_55_cast_fp16)[name = string("op_1395_cast_fp16")]; + tensor var_1397_begin_0 = const()[name = string("op_1397_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1397_end_0 = const()[name = string("op_1397_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_1397_end_mask_0 = const()[name = string("op_1397_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1397_squeeze_mask_0 = const()[name = string("op_1397_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1397_cast_fp16 = slice_by_index(begin = var_1397_begin_0, end = var_1397_end_0, end_mask = var_1397_end_mask_0, squeeze_mask = var_1397_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1397_cast_fp16")]; + bool var_1398_transpose_x_0 = const()[name = string("op_1398_transpose_x_0"), val = bool(false)]; + bool var_1398_transpose_y_0 = const()[name = string("op_1398_transpose_y_0"), val = bool(false)]; + tensor var_1398_cast_fp16 = matmul(transpose_x = var_1398_transpose_x_0, transpose_y = var_1398_transpose_y_0, x = var_1397_cast_fp16, y = v_new_55_cast_fp16)[name = string("op_1398_cast_fp16")]; + tensor var_1399_cast_fp16 = add(x = var_1395_cast_fp16, y = var_1398_cast_fp16)[name = string("op_1399_cast_fp16")]; + tensor var_1401_begin_0 = const()[name = string("op_1401_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1401_end_0 = const()[name = string("op_1401_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_1401_end_mask_0 = const()[name = string("op_1401_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1401_squeeze_mask_0 = const()[name = string("op_1401_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1401_cast_fp16 = slice_by_index(begin = var_1401_begin_0, end = var_1401_end_0, end_mask = var_1401_end_mask_0, squeeze_mask = var_1401_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1401_cast_fp16")]; + tensor var_1402_cast_fp16 = mul(x = state_55_cast_fp16, y = var_1401_cast_fp16)[name = string("op_1402_cast_fp16")]; + tensor var_1404_begin_0 = const()[name = string("op_1404_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_1404_end_0 = const()[name = string("op_1404_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_1404_end_mask_0 = const()[name = string("op_1404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1404_squeeze_mask_0 = const()[name = string("op_1404_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1404_cast_fp16 = slice_by_index(begin = var_1404_begin_0, end = var_1404_end_0, end_mask = var_1404_end_mask_0, squeeze_mask = var_1404_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1404_cast_fp16")]; + bool var_1406_transpose_x_1 = const()[name = string("op_1406_transpose_x_1"), val = bool(true)]; + bool var_1406_transpose_y_1 = const()[name = string("op_1406_transpose_y_1"), val = bool(false)]; + tensor var_1406_cast_fp16 = matmul(transpose_x = var_1406_transpose_x_1, transpose_y = var_1406_transpose_y_1, x = var_1404_cast_fp16, y = v_new_55_cast_fp16)[name = string("op_1406_cast_fp16")]; + tensor state_57_cast_fp16 = add(x = var_1402_cast_fp16, y = var_1406_cast_fp16)[name = string("state_57_cast_fp16")]; + tensor var_1409_begin_0 = const()[name = string("op_1409_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1409_end_0 = const()[name = string("op_1409_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_1409_end_mask_0 = const()[name = string("op_1409_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1409_squeeze_mask_0 = const()[name = string("op_1409_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1409_cast_fp16 = slice_by_index(begin = var_1409_begin_0, end = var_1409_end_0, end_mask = var_1409_end_mask_0, squeeze_mask = var_1409_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1409_cast_fp16")]; + tensor var_1411_begin_0 = const()[name = string("op_1411_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1411_end_0 = const()[name = string("op_1411_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_1411_end_mask_0 = const()[name = string("op_1411_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1411_squeeze_mask_0 = const()[name = string("op_1411_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1411_cast_fp16 = slice_by_index(begin = var_1411_begin_0, end = var_1411_end_0, end_mask = var_1411_end_mask_0, squeeze_mask = var_1411_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1411_cast_fp16")]; + bool var_1412_transpose_x_0 = const()[name = string("op_1412_transpose_x_0"), val = bool(false)]; + bool var_1412_transpose_y_0 = const()[name = string("op_1412_transpose_y_0"), val = bool(false)]; + tensor var_1412_cast_fp16 = matmul(transpose_x = var_1412_transpose_x_0, transpose_y = var_1412_transpose_y_0, x = var_1411_cast_fp16, y = state_57_cast_fp16)[name = string("op_1412_cast_fp16")]; + tensor v_new_57_cast_fp16 = sub(x = var_1409_cast_fp16, y = var_1412_cast_fp16)[name = string("v_new_57_cast_fp16")]; + tensor var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1415_squeeze_mask_0 = const()[name = string("op_1415_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, squeeze_mask = var_1415_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1415_cast_fp16")]; + bool var_1416_transpose_x_0 = const()[name = string("op_1416_transpose_x_0"), val = bool(false)]; + bool var_1416_transpose_y_0 = const()[name = string("op_1416_transpose_y_0"), val = bool(false)]; + tensor var_1416_cast_fp16 = matmul(transpose_x = var_1416_transpose_x_0, transpose_y = var_1416_transpose_y_0, x = var_1415_cast_fp16, y = state_57_cast_fp16)[name = string("op_1416_cast_fp16")]; + tensor var_1418_begin_0 = const()[name = string("op_1418_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1418_end_0 = const()[name = string("op_1418_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_1418_end_mask_0 = const()[name = string("op_1418_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1418_squeeze_mask_0 = const()[name = string("op_1418_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1418_cast_fp16 = slice_by_index(begin = var_1418_begin_0, end = var_1418_end_0, end_mask = var_1418_end_mask_0, squeeze_mask = var_1418_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1418_cast_fp16")]; + bool var_1419_transpose_x_0 = const()[name = string("op_1419_transpose_x_0"), val = bool(false)]; + bool var_1419_transpose_y_0 = const()[name = string("op_1419_transpose_y_0"), val = bool(false)]; + tensor var_1419_cast_fp16 = matmul(transpose_x = var_1419_transpose_x_0, transpose_y = var_1419_transpose_y_0, x = var_1418_cast_fp16, y = v_new_57_cast_fp16)[name = string("op_1419_cast_fp16")]; + tensor var_1420_cast_fp16 = add(x = var_1416_cast_fp16, y = var_1419_cast_fp16)[name = string("op_1420_cast_fp16")]; + tensor var_1422_begin_0 = const()[name = string("op_1422_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1422_end_0 = const()[name = string("op_1422_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_1422_end_mask_0 = const()[name = string("op_1422_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1422_squeeze_mask_0 = const()[name = string("op_1422_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1422_cast_fp16 = slice_by_index(begin = var_1422_begin_0, end = var_1422_end_0, end_mask = var_1422_end_mask_0, squeeze_mask = var_1422_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1422_cast_fp16")]; + tensor var_1423_cast_fp16 = mul(x = state_57_cast_fp16, y = var_1422_cast_fp16)[name = string("op_1423_cast_fp16")]; + tensor var_1425_begin_0 = const()[name = string("op_1425_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_1425_end_0 = const()[name = string("op_1425_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_1425_end_mask_0 = const()[name = string("op_1425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1425_squeeze_mask_0 = const()[name = string("op_1425_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1425_cast_fp16 = slice_by_index(begin = var_1425_begin_0, end = var_1425_end_0, end_mask = var_1425_end_mask_0, squeeze_mask = var_1425_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1425_cast_fp16")]; + bool var_1427_transpose_x_1 = const()[name = string("op_1427_transpose_x_1"), val = bool(true)]; + bool var_1427_transpose_y_1 = const()[name = string("op_1427_transpose_y_1"), val = bool(false)]; + tensor var_1427_cast_fp16 = matmul(transpose_x = var_1427_transpose_x_1, transpose_y = var_1427_transpose_y_1, x = var_1425_cast_fp16, y = v_new_57_cast_fp16)[name = string("op_1427_cast_fp16")]; + tensor state_59_cast_fp16 = add(x = var_1423_cast_fp16, y = var_1427_cast_fp16)[name = string("state_59_cast_fp16")]; + tensor var_1430_begin_0 = const()[name = string("op_1430_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1430_end_0 = const()[name = string("op_1430_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_1430_end_mask_0 = const()[name = string("op_1430_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1430_squeeze_mask_0 = const()[name = string("op_1430_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1430_cast_fp16 = slice_by_index(begin = var_1430_begin_0, end = var_1430_end_0, end_mask = var_1430_end_mask_0, squeeze_mask = var_1430_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1430_cast_fp16")]; + tensor var_1432_begin_0 = const()[name = string("op_1432_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1432_end_0 = const()[name = string("op_1432_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_1432_end_mask_0 = const()[name = string("op_1432_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1432_squeeze_mask_0 = const()[name = string("op_1432_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1432_cast_fp16 = slice_by_index(begin = var_1432_begin_0, end = var_1432_end_0, end_mask = var_1432_end_mask_0, squeeze_mask = var_1432_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1432_cast_fp16")]; + bool var_1433_transpose_x_0 = const()[name = string("op_1433_transpose_x_0"), val = bool(false)]; + bool var_1433_transpose_y_0 = const()[name = string("op_1433_transpose_y_0"), val = bool(false)]; + tensor var_1433_cast_fp16 = matmul(transpose_x = var_1433_transpose_x_0, transpose_y = var_1433_transpose_y_0, x = var_1432_cast_fp16, y = state_59_cast_fp16)[name = string("op_1433_cast_fp16")]; + tensor v_new_59_cast_fp16 = sub(x = var_1430_cast_fp16, y = var_1433_cast_fp16)[name = string("v_new_59_cast_fp16")]; + tensor var_1436_begin_0 = const()[name = string("op_1436_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1436_end_0 = const()[name = string("op_1436_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_1436_end_mask_0 = const()[name = string("op_1436_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1436_squeeze_mask_0 = const()[name = string("op_1436_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1436_cast_fp16 = slice_by_index(begin = var_1436_begin_0, end = var_1436_end_0, end_mask = var_1436_end_mask_0, squeeze_mask = var_1436_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1436_cast_fp16")]; + bool var_1437_transpose_x_0 = const()[name = string("op_1437_transpose_x_0"), val = bool(false)]; + bool var_1437_transpose_y_0 = const()[name = string("op_1437_transpose_y_0"), val = bool(false)]; + tensor var_1437_cast_fp16 = matmul(transpose_x = var_1437_transpose_x_0, transpose_y = var_1437_transpose_y_0, x = var_1436_cast_fp16, y = state_59_cast_fp16)[name = string("op_1437_cast_fp16")]; + tensor var_1439_begin_0 = const()[name = string("op_1439_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1439_end_0 = const()[name = string("op_1439_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_1439_end_mask_0 = const()[name = string("op_1439_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1439_squeeze_mask_0 = const()[name = string("op_1439_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1439_cast_fp16 = slice_by_index(begin = var_1439_begin_0, end = var_1439_end_0, end_mask = var_1439_end_mask_0, squeeze_mask = var_1439_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1439_cast_fp16")]; + bool var_1440_transpose_x_0 = const()[name = string("op_1440_transpose_x_0"), val = bool(false)]; + bool var_1440_transpose_y_0 = const()[name = string("op_1440_transpose_y_0"), val = bool(false)]; + tensor var_1440_cast_fp16 = matmul(transpose_x = var_1440_transpose_x_0, transpose_y = var_1440_transpose_y_0, x = var_1439_cast_fp16, y = v_new_59_cast_fp16)[name = string("op_1440_cast_fp16")]; + tensor var_1441_cast_fp16 = add(x = var_1437_cast_fp16, y = var_1440_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor var_1443_begin_0 = const()[name = string("op_1443_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1443_end_0 = const()[name = string("op_1443_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_1443_end_mask_0 = const()[name = string("op_1443_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1443_squeeze_mask_0 = const()[name = string("op_1443_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1443_cast_fp16 = slice_by_index(begin = var_1443_begin_0, end = var_1443_end_0, end_mask = var_1443_end_mask_0, squeeze_mask = var_1443_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1443_cast_fp16")]; + tensor var_1444_cast_fp16 = mul(x = state_59_cast_fp16, y = var_1443_cast_fp16)[name = string("op_1444_cast_fp16")]; + tensor var_1446_begin_0 = const()[name = string("op_1446_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_1446_end_0 = const()[name = string("op_1446_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_1446_end_mask_0 = const()[name = string("op_1446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1446_squeeze_mask_0 = const()[name = string("op_1446_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1446_cast_fp16 = slice_by_index(begin = var_1446_begin_0, end = var_1446_end_0, end_mask = var_1446_end_mask_0, squeeze_mask = var_1446_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1446_cast_fp16")]; + bool var_1448_transpose_x_1 = const()[name = string("op_1448_transpose_x_1"), val = bool(true)]; + bool var_1448_transpose_y_1 = const()[name = string("op_1448_transpose_y_1"), val = bool(false)]; + tensor var_1448_cast_fp16 = matmul(transpose_x = var_1448_transpose_x_1, transpose_y = var_1448_transpose_y_1, x = var_1446_cast_fp16, y = v_new_59_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor state_61_cast_fp16 = add(x = var_1444_cast_fp16, y = var_1448_cast_fp16)[name = string("state_61_cast_fp16")]; + tensor var_1451_begin_0 = const()[name = string("op_1451_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1451_end_0 = const()[name = string("op_1451_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_1451_end_mask_0 = const()[name = string("op_1451_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1451_squeeze_mask_0 = const()[name = string("op_1451_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1451_cast_fp16 = slice_by_index(begin = var_1451_begin_0, end = var_1451_end_0, end_mask = var_1451_end_mask_0, squeeze_mask = var_1451_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1451_cast_fp16")]; + tensor var_1453_begin_0 = const()[name = string("op_1453_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1453_end_0 = const()[name = string("op_1453_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_1453_end_mask_0 = const()[name = string("op_1453_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1453_squeeze_mask_0 = const()[name = string("op_1453_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1453_cast_fp16 = slice_by_index(begin = var_1453_begin_0, end = var_1453_end_0, end_mask = var_1453_end_mask_0, squeeze_mask = var_1453_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1453_cast_fp16")]; + bool var_1454_transpose_x_0 = const()[name = string("op_1454_transpose_x_0"), val = bool(false)]; + bool var_1454_transpose_y_0 = const()[name = string("op_1454_transpose_y_0"), val = bool(false)]; + tensor var_1454_cast_fp16 = matmul(transpose_x = var_1454_transpose_x_0, transpose_y = var_1454_transpose_y_0, x = var_1453_cast_fp16, y = state_61_cast_fp16)[name = string("op_1454_cast_fp16")]; + tensor v_new_61_cast_fp16 = sub(x = var_1451_cast_fp16, y = var_1454_cast_fp16)[name = string("v_new_61_cast_fp16")]; + tensor var_1457_begin_0 = const()[name = string("op_1457_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1457_end_0 = const()[name = string("op_1457_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_1457_end_mask_0 = const()[name = string("op_1457_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1457_squeeze_mask_0 = const()[name = string("op_1457_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1457_cast_fp16 = slice_by_index(begin = var_1457_begin_0, end = var_1457_end_0, end_mask = var_1457_end_mask_0, squeeze_mask = var_1457_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1457_cast_fp16")]; + bool var_1458_transpose_x_0 = const()[name = string("op_1458_transpose_x_0"), val = bool(false)]; + bool var_1458_transpose_y_0 = const()[name = string("op_1458_transpose_y_0"), val = bool(false)]; + tensor var_1458_cast_fp16 = matmul(transpose_x = var_1458_transpose_x_0, transpose_y = var_1458_transpose_y_0, x = var_1457_cast_fp16, y = state_61_cast_fp16)[name = string("op_1458_cast_fp16")]; + tensor var_1460_begin_0 = const()[name = string("op_1460_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1460_end_0 = const()[name = string("op_1460_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_1460_end_mask_0 = const()[name = string("op_1460_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1460_squeeze_mask_0 = const()[name = string("op_1460_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1460_cast_fp16 = slice_by_index(begin = var_1460_begin_0, end = var_1460_end_0, end_mask = var_1460_end_mask_0, squeeze_mask = var_1460_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1460_cast_fp16")]; + bool var_1461_transpose_x_0 = const()[name = string("op_1461_transpose_x_0"), val = bool(false)]; + bool var_1461_transpose_y_0 = const()[name = string("op_1461_transpose_y_0"), val = bool(false)]; + tensor var_1461_cast_fp16 = matmul(transpose_x = var_1461_transpose_x_0, transpose_y = var_1461_transpose_y_0, x = var_1460_cast_fp16, y = v_new_61_cast_fp16)[name = string("op_1461_cast_fp16")]; + tensor var_1462_cast_fp16 = add(x = var_1458_cast_fp16, y = var_1461_cast_fp16)[name = string("op_1462_cast_fp16")]; + tensor var_1464_begin_0 = const()[name = string("op_1464_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1464_end_0 = const()[name = string("op_1464_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_1464_end_mask_0 = const()[name = string("op_1464_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1464_squeeze_mask_0 = const()[name = string("op_1464_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1464_cast_fp16 = slice_by_index(begin = var_1464_begin_0, end = var_1464_end_0, end_mask = var_1464_end_mask_0, squeeze_mask = var_1464_squeeze_mask_0, x = chunk_decay_3_cast_fp16)[name = string("op_1464_cast_fp16")]; + tensor var_1465_cast_fp16 = mul(x = state_61_cast_fp16, y = var_1464_cast_fp16)[name = string("op_1465_cast_fp16")]; + tensor var_1467_begin_0 = const()[name = string("op_1467_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_1467_end_0 = const()[name = string("op_1467_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_1467_end_mask_0 = const()[name = string("op_1467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1467_squeeze_mask_0 = const()[name = string("op_1467_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1467_cast_fp16 = slice_by_index(begin = var_1467_begin_0, end = var_1467_end_0, end_mask = var_1467_end_mask_0, squeeze_mask = var_1467_squeeze_mask_0, x = k_dec_3_cast_fp16)[name = string("op_1467_cast_fp16")]; + bool var_1469_transpose_x_1 = const()[name = string("op_1469_transpose_x_1"), val = bool(true)]; + bool var_1469_transpose_y_1 = const()[name = string("op_1469_transpose_y_1"), val = bool(false)]; + tensor var_1469_cast_fp16 = matmul(transpose_x = var_1469_transpose_x_1, transpose_y = var_1469_transpose_y_1, x = var_1467_cast_fp16, y = v_new_61_cast_fp16)[name = string("op_1469_cast_fp16")]; + tensor state_63_cast_fp16 = add(x = var_1465_cast_fp16, y = var_1469_cast_fp16)[name = string("state_63_cast_fp16")]; + tensor var_1472_begin_0 = const()[name = string("op_1472_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_1472_end_0 = const()[name = string("op_1472_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_1472_end_mask_0 = const()[name = string("op_1472_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1472_squeeze_mask_0 = const()[name = string("op_1472_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1472_cast_fp16 = slice_by_index(begin = var_1472_begin_0, end = var_1472_end_0, end_mask = var_1472_end_mask_0, squeeze_mask = var_1472_squeeze_mask_0, x = new_v_3_cast_fp16)[name = string("op_1472_cast_fp16")]; + tensor var_1474_begin_0 = const()[name = string("op_1474_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_1474_end_0 = const()[name = string("op_1474_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_1474_end_mask_0 = const()[name = string("op_1474_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1474_squeeze_mask_0 = const()[name = string("op_1474_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1474_cast_fp16 = slice_by_index(begin = var_1474_begin_0, end = var_1474_end_0, end_mask = var_1474_end_mask_0, squeeze_mask = var_1474_squeeze_mask_0, x = k_cumdecay_3_cast_fp16)[name = string("op_1474_cast_fp16")]; + bool var_1475_transpose_x_0 = const()[name = string("op_1475_transpose_x_0"), val = bool(false)]; + bool var_1475_transpose_y_0 = const()[name = string("op_1475_transpose_y_0"), val = bool(false)]; + tensor var_1475_cast_fp16 = matmul(transpose_x = var_1475_transpose_x_0, transpose_y = var_1475_transpose_y_0, x = var_1474_cast_fp16, y = state_63_cast_fp16)[name = string("op_1475_cast_fp16")]; + tensor v_new_63_cast_fp16 = sub(x = var_1472_cast_fp16, y = var_1475_cast_fp16)[name = string("v_new_63_cast_fp16")]; + tensor var_1478_begin_0 = const()[name = string("op_1478_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_1478_end_0 = const()[name = string("op_1478_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_1478_end_mask_0 = const()[name = string("op_1478_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1478_squeeze_mask_0 = const()[name = string("op_1478_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1478_cast_fp16 = slice_by_index(begin = var_1478_begin_0, end = var_1478_end_0, end_mask = var_1478_end_mask_0, squeeze_mask = var_1478_squeeze_mask_0, x = q_dec_3_cast_fp16)[name = string("op_1478_cast_fp16")]; + bool var_1479_transpose_x_0 = const()[name = string("op_1479_transpose_x_0"), val = bool(false)]; + bool var_1479_transpose_y_0 = const()[name = string("op_1479_transpose_y_0"), val = bool(false)]; + tensor var_1479_cast_fp16 = matmul(transpose_x = var_1479_transpose_x_0, transpose_y = var_1479_transpose_y_0, x = var_1478_cast_fp16, y = state_63_cast_fp16)[name = string("op_1479_cast_fp16")]; + tensor var_1481_begin_0 = const()[name = string("op_1481_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_1481_end_0 = const()[name = string("op_1481_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_1481_end_mask_0 = const()[name = string("op_1481_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1481_squeeze_mask_0 = const()[name = string("op_1481_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1481_cast_fp16 = slice_by_index(begin = var_1481_begin_0, end = var_1481_end_0, end_mask = var_1481_end_mask_0, squeeze_mask = var_1481_squeeze_mask_0, x = attn_intra_3_cast_fp16)[name = string("op_1481_cast_fp16")]; + bool var_1482_transpose_x_0 = const()[name = string("op_1482_transpose_x_0"), val = bool(false)]; + bool var_1482_transpose_y_0 = const()[name = string("op_1482_transpose_y_0"), val = bool(false)]; + tensor var_1482_cast_fp16 = matmul(transpose_x = var_1482_transpose_x_0, transpose_y = var_1482_transpose_y_0, x = var_1481_cast_fp16, y = v_new_63_cast_fp16)[name = string("op_1482_cast_fp16")]; + tensor var_1483_cast_fp16 = add(x = var_1479_cast_fp16, y = var_1482_cast_fp16)[name = string("op_1483_cast_fp16")]; + int32 var_1485_axis_0 = const()[name = string("op_1485_axis_0"), val = int32(1)]; + tensor var_1485_cast_fp16 = stack(axis = var_1485_axis_0, values = (var_1168_cast_fp16, var_1189_cast_fp16, var_1210_cast_fp16, var_1231_cast_fp16, var_1252_cast_fp16, var_1273_cast_fp16, var_1294_cast_fp16, var_1315_cast_fp16, var_1336_cast_fp16, var_1357_cast_fp16, var_1378_cast_fp16, var_1399_cast_fp16, var_1420_cast_fp16, var_1441_cast_fp16, var_1462_cast_fp16, var_1483_cast_fp16))[name = string("op_1485_cast_fp16")]; + tensor var_1486 = const()[name = string("op_1486"), val = tensor([32, 1024, 128])]; + tensor var_1487_cast_fp16 = reshape(shape = var_1486, x = var_1485_cast_fp16)[name = string("op_1487_cast_fp16")]; + tensor var_1488 = const()[name = string("op_1488"), val = tensor([1, 0, 2])]; + tensor input_23_begin_0 = const()[name = string("input_23_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_23_end_0 = const()[name = string("input_23_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_23_end_mask_0 = const()[name = string("input_23_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_23_squeeze_mask_0 = const()[name = string("input_23_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_23_cast_fp16 = slice_by_index(begin = input_23_begin_0, end = input_23_end_0, end_mask = input_23_end_mask_0, squeeze_mask = input_23_squeeze_mask_0, x = z_3_cast_fp16)[name = string("input_23_cast_fp16")]; + fp16 var_810_promoted_1_to_fp16 = const()[name = string("op_810_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_43_cast_fp16 = transpose(perm = var_1488, x = var_1487_cast_fp16)[name = string("transpose_40")]; + tensor var_1492_cast_fp16 = pow(x = x_43_cast_fp16, y = var_810_promoted_1_to_fp16)[name = string("op_1492_cast_fp16")]; + tensor var_1494_axes_0 = const()[name = string("op_1494_axes_0"), val = tensor([-1])]; + bool var_1494_keep_dims_0 = const()[name = string("op_1494_keep_dims_0"), val = bool(true)]; + tensor var_1494_cast_fp16 = reduce_mean(axes = var_1494_axes_0, keep_dims = var_1494_keep_dims_0, x = var_1492_cast_fp16)[name = string("op_1494_cast_fp16")]; + fp16 var_1495_to_fp16 = const()[name = string("op_1495_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_1496_cast_fp16 = add(x = var_1494_cast_fp16, y = var_1495_to_fp16)[name = string("op_1496_cast_fp16")]; + fp32 var_1497_epsilon_0 = const()[name = string("op_1497_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1497_cast_fp16 = rsqrt(epsilon = var_1497_epsilon_0, x = var_1496_cast_fp16)[name = string("op_1497_cast_fp16")]; + tensor x_45_cast_fp16 = mul(x = x_43_cast_fp16, y = var_1497_cast_fp16)[name = string("x_45_cast_fp16")]; + tensor layers_1_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_1_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292950528)))]; + tensor var_1499_cast_fp16 = mul(x = layers_1_linear_attn_norm_weight_to_fp16, y = x_45_cast_fp16)[name = string("op_1499_cast_fp16")]; + tensor var_1500_cast_fp16 = silu(x = input_23_cast_fp16)[name = string("op_1500_cast_fp16")]; + tensor core_3_cast_fp16 = mul(x = var_1499_cast_fp16, y = var_1500_cast_fp16)[name = string("core_3_cast_fp16")]; + tensor var_1502 = const()[name = string("op_1502"), val = tensor([1, 1024, 4096])]; + tensor input_25_cast_fp16 = reshape(shape = var_1502, x = core_3_cast_fp16)[name = string("input_25_cast_fp16")]; + tensor layers_1_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_1_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292950848)))]; + tensor linear_12_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_1_linear_attn_out_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_12_cast_fp16")]; + tensor x_47_cast_fp16 = add(x = x_25_cast_fp16, y = linear_12_cast_fp16)[name = string("x_47_cast_fp16")]; + fp16 var_810_promoted_2_to_fp16 = const()[name = string("op_810_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_1508_cast_fp16 = pow(x = x_47_cast_fp16, y = var_810_promoted_2_to_fp16)[name = string("op_1508_cast_fp16")]; + tensor var_1510_axes_0 = const()[name = string("op_1510_axes_0"), val = tensor([-1])]; + bool var_1510_keep_dims_0 = const()[name = string("op_1510_keep_dims_0"), val = bool(true)]; + tensor var_1510_cast_fp16 = reduce_mean(axes = var_1510_axes_0, keep_dims = var_1510_keep_dims_0, x = var_1508_cast_fp16)[name = string("op_1510_cast_fp16")]; + fp16 var_1511_to_fp16 = const()[name = string("op_1511_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_1512_cast_fp16 = add(x = var_1510_cast_fp16, y = var_1511_to_fp16)[name = string("op_1512_cast_fp16")]; + fp32 var_1513_epsilon_0 = const()[name = string("op_1513_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1513_cast_fp16 = rsqrt(epsilon = var_1513_epsilon_0, x = var_1512_cast_fp16)[name = string("op_1513_cast_fp16")]; + tensor x_49_cast_fp16 = mul(x = x_47_cast_fp16, y = var_1513_cast_fp16)[name = string("x_49_cast_fp16")]; + tensor var_1516_to_fp16 = const()[name = string("op_1516_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313922432)))]; + tensor input_27_cast_fp16 = mul(x = x_49_cast_fp16, y = var_1516_to_fp16)[name = string("input_27_cast_fp16")]; + tensor layers_1_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313927616)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_1_mlp_gate_proj_weight_to_fp16, x = input_27_cast_fp16)[name = string("linear_13_cast_fp16")]; + tensor var_1523_cast_fp16 = silu(x = linear_13_cast_fp16)[name = string("op_1523_cast_fp16")]; + tensor layers_1_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361113600)))]; + tensor linear_14_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_1_mlp_up_proj_weight_to_fp16, x = input_27_cast_fp16)[name = string("linear_14_cast_fp16")]; + tensor input_31_cast_fp16 = mul(x = var_1523_cast_fp16, y = linear_14_cast_fp16)[name = string("input_31_cast_fp16")]; + tensor layers_1_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408299584)))]; + tensor linear_15_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_1_mlp_down_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_15_cast_fp16")]; + tensor x_51_cast_fp16 = add(x = x_47_cast_fp16, y = linear_15_cast_fp16)[name = string("x_51_cast_fp16")]; + int32 var_1543 = const()[name = string("op_1543"), val = int32(-2)]; + int32 var_1564 = const()[name = string("op_1564"), val = int32(-1)]; + fp16 var_1563_promoted_to_fp16 = const()[name = string("op_1563_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_1573_cast_fp16 = pow(x = x_51_cast_fp16, y = var_1563_promoted_to_fp16)[name = string("op_1573_cast_fp16")]; + tensor var_1575_axes_0 = const()[name = string("op_1575_axes_0"), val = tensor([-1])]; + bool var_1575_keep_dims_0 = const()[name = string("op_1575_keep_dims_0"), val = bool(true)]; + tensor var_1575_cast_fp16 = reduce_mean(axes = var_1575_axes_0, keep_dims = var_1575_keep_dims_0, x = var_1573_cast_fp16)[name = string("op_1575_cast_fp16")]; + fp16 var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_1577_cast_fp16 = add(x = var_1575_cast_fp16, y = var_1576_to_fp16)[name = string("op_1577_cast_fp16")]; + fp32 var_1578_epsilon_0 = const()[name = string("op_1578_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1578_cast_fp16 = rsqrt(epsilon = var_1578_epsilon_0, x = var_1577_cast_fp16)[name = string("op_1578_cast_fp16")]; + tensor x_53_cast_fp16 = mul(x = x_51_cast_fp16, y = var_1578_cast_fp16)[name = string("x_53_cast_fp16")]; + tensor var_1581_to_fp16 = const()[name = string("op_1581_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(455485568)))]; + tensor x_55_cast_fp16 = mul(x = x_53_cast_fp16, y = var_1581_to_fp16)[name = string("x_55_cast_fp16")]; + tensor layers_2_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_2_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(455490752)))]; + tensor linear_16_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_linear_attn_in_proj_qkv_weight_to_fp16, x = x_55_cast_fp16)[name = string("linear_16_cast_fp16")]; + tensor input_33_perm_0 = const()[name = string("input_33_perm_0"), val = tensor([0, 2, 1])]; + tensor input_35_pad_0 = const()[name = string("input_35_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("constant")]; + fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)]; + tensor input_33_cast_fp16 = transpose(perm = input_33_perm_0, x = linear_16_cast_fp16)[name = string("transpose_39")]; + tensor input_35_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = input_35_mode_0, pad = input_35_pad_0, x = input_33_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(8192)]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1])]; + tensor layers_2_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_2_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497433856)))]; + tensor input_37_cast_fp16 = conv(dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_2_linear_attn_conv1d_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + tensor var_1610_cast_fp16 = silu(x = input_37_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor qkv_5_perm_0 = const()[name = string("qkv_5_perm_0"), val = tensor([0, 2, 1])]; + tensor var_1612_begin_0 = const()[name = string("op_1612_begin_0"), val = tensor([0, 0, 0])]; + tensor var_1612_end_0 = const()[name = string("op_1612_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_1612_end_mask_0 = const()[name = string("op_1612_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_5_cast_fp16 = transpose(perm = qkv_5_perm_0, x = var_1610_cast_fp16)[name = string("transpose_38")]; + tensor var_1612_cast_fp16 = slice_by_index(begin = var_1612_begin_0, end = var_1612_end_0, end_mask = var_1612_end_mask_0, x = qkv_5_cast_fp16)[name = string("op_1612_cast_fp16")]; + tensor var_1613 = const()[name = string("op_1613"), val = tensor([1, 1024, 16, 128])]; + tensor q_21_cast_fp16 = reshape(shape = var_1613, x = var_1612_cast_fp16)[name = string("q_21_cast_fp16")]; + tensor var_1615_begin_0 = const()[name = string("op_1615_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_1615_end_0 = const()[name = string("op_1615_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_1615_end_mask_0 = const()[name = string("op_1615_end_mask_0"), val = tensor([true, true, false])]; + tensor var_1615_cast_fp16 = slice_by_index(begin = var_1615_begin_0, end = var_1615_end_0, end_mask = var_1615_end_mask_0, x = qkv_5_cast_fp16)[name = string("op_1615_cast_fp16")]; + tensor var_1616 = const()[name = string("op_1616"), val = tensor([1, 1024, 16, 128])]; + tensor k_17_cast_fp16 = reshape(shape = var_1616, x = var_1615_cast_fp16)[name = string("k_17_cast_fp16")]; + tensor var_1618_begin_0 = const()[name = string("op_1618_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_1618_end_0 = const()[name = string("op_1618_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_1618_end_mask_0 = const()[name = string("op_1618_end_mask_0"), val = tensor([true, true, true])]; + tensor var_1618_cast_fp16 = slice_by_index(begin = var_1618_begin_0, end = var_1618_end_0, end_mask = var_1618_end_mask_0, x = qkv_5_cast_fp16)[name = string("op_1618_cast_fp16")]; + tensor var_1619 = const()[name = string("op_1619"), val = tensor([1, 1024, 32, 128])]; + tensor v_9_cast_fp16 = reshape(shape = var_1619, x = var_1618_cast_fp16)[name = string("v_9_cast_fp16")]; + tensor layers_2_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_2_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497499456)))]; + tensor linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_2_linear_attn_in_proj_z_weight_to_fp16, x = x_55_cast_fp16)[name = string("linear_17_cast_fp16")]; + tensor var_1623 = const()[name = string("op_1623"), val = tensor([1, 1024, 32, 128])]; + tensor z_5_cast_fp16 = reshape(shape = var_1623, x = linear_17_cast_fp16)[name = string("z_5_cast_fp16")]; + tensor layers_2_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_2_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518471040)))]; + tensor linear_18_bias_0_to_fp16 = const()[name = string("linear_18_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518634944)))]; + tensor linear_18_cast_fp16 = linear(bias = linear_18_bias_0_to_fp16, weight = layers_2_linear_attn_in_proj_b_weight_to_fp16, x = x_55_cast_fp16)[name = string("linear_18_cast_fp16")]; + tensor beta_9_cast_fp16 = sigmoid(x = linear_18_cast_fp16)[name = string("beta_9_cast_fp16")]; + tensor var_1632_weight_0_to_fp16 = const()[name = string("op_1632_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518635072)))]; + tensor var_1632_bias_0_to_fp16 = const()[name = string("op_1632_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518798976)))]; + tensor var_1632_cast_fp16 = linear(bias = var_1632_bias_0_to_fp16, weight = var_1632_weight_0_to_fp16, x = x_55_cast_fp16)[name = string("op_1632_cast_fp16")]; + tensor var_1633_cast_fp16 = softplus(x = var_1632_cast_fp16)[name = string("op_1633_cast_fp16")]; + tensor var_1629_to_fp16 = const()[name = string("op_1629_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518799104)))]; + tensor g_9_cast_fp16 = mul(x = var_1629_to_fp16, y = var_1633_cast_fp16)[name = string("g_9_cast_fp16")]; + tensor var_1635_cast_fp16 = mul(x = q_21_cast_fp16, y = q_21_cast_fp16)[name = string("op_1635_cast_fp16")]; + tensor var_1637_axes_0 = const()[name = string("op_1637_axes_0"), val = tensor([-1])]; + bool var_1637_keep_dims_0 = const()[name = string("op_1637_keep_dims_0"), val = bool(true)]; + tensor var_1637_cast_fp16 = reduce_sum(axes = var_1637_axes_0, keep_dims = var_1637_keep_dims_0, x = var_1635_cast_fp16)[name = string("op_1637_cast_fp16")]; + fp16 var_1638_to_fp16 = const()[name = string("op_1638_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_1639_cast_fp16 = add(x = var_1637_cast_fp16, y = var_1638_to_fp16)[name = string("op_1639_cast_fp16")]; + fp32 var_1640_epsilon_0 = const()[name = string("op_1640_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1640_cast_fp16 = rsqrt(epsilon = var_1640_epsilon_0, x = var_1639_cast_fp16)[name = string("op_1640_cast_fp16")]; + tensor q_23_cast_fp16 = mul(x = q_21_cast_fp16, y = var_1640_cast_fp16)[name = string("q_23_cast_fp16")]; + tensor var_1642_cast_fp16 = mul(x = k_17_cast_fp16, y = k_17_cast_fp16)[name = string("op_1642_cast_fp16")]; + tensor var_1644_axes_0 = const()[name = string("op_1644_axes_0"), val = tensor([-1])]; + bool var_1644_keep_dims_0 = const()[name = string("op_1644_keep_dims_0"), val = bool(true)]; + tensor var_1644_cast_fp16 = reduce_sum(axes = var_1644_axes_0, keep_dims = var_1644_keep_dims_0, x = var_1642_cast_fp16)[name = string("op_1644_cast_fp16")]; + fp16 var_1645_to_fp16 = const()[name = string("op_1645_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_1646_cast_fp16 = add(x = var_1644_cast_fp16, y = var_1645_to_fp16)[name = string("op_1646_cast_fp16")]; + fp32 var_1647_epsilon_0 = const()[name = string("op_1647_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_1647_cast_fp16 = rsqrt(epsilon = var_1647_epsilon_0, x = var_1646_cast_fp16)[name = string("op_1647_cast_fp16")]; + tensor k_19_cast_fp16 = mul(x = k_17_cast_fp16, y = var_1647_cast_fp16)[name = string("k_19_cast_fp16")]; + tensor var_1652_axes_0 = const()[name = string("op_1652_axes_0"), val = tensor([3])]; + tensor var_1652_cast_fp16 = expand_dims(axes = var_1652_axes_0, x = q_23_cast_fp16)[name = string("op_1652_cast_fp16")]; + tensor var_1654_reps_0 = const()[name = string("op_1654_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_1654_cast_fp16 = tile(reps = var_1654_reps_0, x = var_1652_cast_fp16)[name = string("op_1654_cast_fp16")]; + tensor var_1655 = const()[name = string("op_1655"), val = tensor([1, 1024, 32, 128])]; + tensor q_25_cast_fp16 = reshape(shape = var_1655, x = var_1654_cast_fp16)[name = string("q_25_cast_fp16")]; + tensor var_1660_axes_0 = const()[name = string("op_1660_axes_0"), val = tensor([3])]; + tensor var_1660_cast_fp16 = expand_dims(axes = var_1660_axes_0, x = k_19_cast_fp16)[name = string("op_1660_cast_fp16")]; + tensor var_1662_reps_0 = const()[name = string("op_1662_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_1662_cast_fp16 = tile(reps = var_1662_reps_0, x = var_1660_cast_fp16)[name = string("op_1662_cast_fp16")]; + tensor var_1663 = const()[name = string("op_1663"), val = tensor([1, 1024, 32, 128])]; + tensor k_21_cast_fp16 = reshape(shape = var_1663, x = var_1662_cast_fp16)[name = string("k_21_cast_fp16")]; + fp16 var_1665_to_fp16 = const()[name = string("op_1665_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_27_cast_fp16 = mul(x = q_25_cast_fp16, y = var_1665_to_fp16)[name = string("q_27_cast_fp16")]; + tensor var_1667_begin_0 = const()[name = string("op_1667_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1667_end_0 = const()[name = string("op_1667_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_1667_end_mask_0 = const()[name = string("op_1667_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_1667_squeeze_mask_0 = const()[name = string("op_1667_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_1667_cast_fp16 = slice_by_index(begin = var_1667_begin_0, end = var_1667_end_0, end_mask = var_1667_end_mask_0, squeeze_mask = var_1667_squeeze_mask_0, x = q_27_cast_fp16)[name = string("op_1667_cast_fp16")]; + tensor var_1668 = const()[name = string("op_1668"), val = tensor([1, 0, 2])]; + tensor var_1670 = const()[name = string("op_1670"), val = tensor([32, 16, 64, 128])]; + tensor var_1669_cast_fp16 = transpose(perm = var_1668, x = var_1667_cast_fp16)[name = string("transpose_37")]; + tensor q_29_cast_fp16 = reshape(shape = var_1670, x = var_1669_cast_fp16)[name = string("q_29_cast_fp16")]; + tensor var_1672_begin_0 = const()[name = string("op_1672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1672_end_0 = const()[name = string("op_1672_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_1672_end_mask_0 = const()[name = string("op_1672_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_1672_squeeze_mask_0 = const()[name = string("op_1672_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_1672_cast_fp16 = slice_by_index(begin = var_1672_begin_0, end = var_1672_end_0, end_mask = var_1672_end_mask_0, squeeze_mask = var_1672_squeeze_mask_0, x = k_21_cast_fp16)[name = string("op_1672_cast_fp16")]; + tensor var_1673 = const()[name = string("op_1673"), val = tensor([1, 0, 2])]; + tensor var_1675 = const()[name = string("op_1675"), val = tensor([32, 16, 64, 128])]; + tensor var_1674_cast_fp16 = transpose(perm = var_1673, x = var_1672_cast_fp16)[name = string("transpose_36")]; + tensor k_23_cast_fp16 = reshape(shape = var_1675, x = var_1674_cast_fp16)[name = string("k_23_cast_fp16")]; + tensor var_1677_begin_0 = const()[name = string("op_1677_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1677_end_0 = const()[name = string("op_1677_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_1677_end_mask_0 = const()[name = string("op_1677_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_1677_squeeze_mask_0 = const()[name = string("op_1677_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_1677_cast_fp16 = slice_by_index(begin = var_1677_begin_0, end = var_1677_end_0, end_mask = var_1677_end_mask_0, squeeze_mask = var_1677_squeeze_mask_0, x = v_9_cast_fp16)[name = string("op_1677_cast_fp16")]; + tensor var_1678 = const()[name = string("op_1678"), val = tensor([1, 0, 2])]; + tensor var_1680 = const()[name = string("op_1680"), val = tensor([32, 16, 64, 128])]; + tensor var_1679_cast_fp16 = transpose(perm = var_1678, x = var_1677_cast_fp16)[name = string("transpose_35")]; + tensor v_11_cast_fp16 = reshape(shape = var_1680, x = var_1679_cast_fp16)[name = string("v_11_cast_fp16")]; + tensor var_1682_begin_0 = const()[name = string("op_1682_begin_0"), val = tensor([0, 0, 0])]; + tensor var_1682_end_0 = const()[name = string("op_1682_end_0"), val = tensor([1, 1024, 32])]; + tensor var_1682_end_mask_0 = const()[name = string("op_1682_end_mask_0"), val = tensor([false, true, true])]; + tensor var_1682_squeeze_mask_0 = const()[name = string("op_1682_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_1682_cast_fp16 = slice_by_index(begin = var_1682_begin_0, end = var_1682_end_0, end_mask = var_1682_end_mask_0, squeeze_mask = var_1682_squeeze_mask_0, x = beta_9_cast_fp16)[name = string("op_1682_cast_fp16")]; + tensor var_1683_perm_0 = const()[name = string("op_1683_perm_0"), val = tensor([1, 0])]; + tensor var_1684 = const()[name = string("op_1684"), val = tensor([32, 16, 64, 1])]; + tensor var_1683_cast_fp16 = transpose(perm = var_1683_perm_0, x = var_1682_cast_fp16)[name = string("transpose_34")]; + tensor beta_11_cast_fp16 = reshape(shape = var_1684, x = var_1683_cast_fp16)[name = string("beta_11_cast_fp16")]; + tensor var_1686_begin_0 = const()[name = string("op_1686_begin_0"), val = tensor([0, 0, 0])]; + tensor var_1686_end_0 = const()[name = string("op_1686_end_0"), val = tensor([1, 1024, 32])]; + tensor var_1686_end_mask_0 = const()[name = string("op_1686_end_mask_0"), val = tensor([false, true, true])]; + tensor var_1686_squeeze_mask_0 = const()[name = string("op_1686_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_1686_cast_fp16 = slice_by_index(begin = var_1686_begin_0, end = var_1686_end_0, end_mask = var_1686_end_mask_0, squeeze_mask = var_1686_squeeze_mask_0, x = g_9_cast_fp16)[name = string("op_1686_cast_fp16")]; + tensor var_1687_perm_0 = const()[name = string("op_1687_perm_0"), val = tensor([1, 0])]; + tensor var_1688 = const()[name = string("op_1688"), val = tensor([32, 16, 64])]; + tensor var_1687_cast_fp16 = transpose(perm = var_1687_perm_0, x = var_1686_cast_fp16)[name = string("transpose_33")]; + tensor g_11_cast_fp16 = reshape(shape = var_1688, x = var_1687_cast_fp16)[name = string("g_11_cast_fp16")]; + tensor g_col_5_axes_0 = const()[name = string("g_col_5_axes_0"), val = tensor([-1])]; + tensor g_col_5_cast_fp16 = expand_dims(axes = g_col_5_axes_0, x = g_11_cast_fp16)[name = string("g_col_5_cast_fp16")]; + bool var_1691_transpose_x_0 = const()[name = string("op_1691_transpose_x_0"), val = bool(false)]; + bool var_1691_transpose_y_0 = const()[name = string("op_1691_transpose_y_0"), val = bool(false)]; + tensor var_1691_cast_fp16 = matmul(transpose_x = var_1691_transpose_x_0, transpose_y = var_1691_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_5_cast_fp16)[name = string("op_1691_cast_fp16")]; + tensor cum_5_axes_0 = const()[name = string("cum_5_axes_0"), val = tensor([-1])]; + tensor cum_5_cast_fp16 = squeeze(axes = cum_5_axes_0, x = var_1691_cast_fp16)[name = string("cum_5_cast_fp16")]; + bool var_1693_transpose_x_0 = const()[name = string("op_1693_transpose_x_0"), val = bool(false)]; + bool var_1693_transpose_y_0 = const()[name = string("op_1693_transpose_y_0"), val = bool(false)]; + tensor var_1693_cast_fp16 = matmul(transpose_x = var_1693_transpose_x_0, transpose_y = var_1693_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_5_cast_fp16)[name = string("op_1693_cast_fp16")]; + tensor to_end_5_axes_0 = const()[name = string("to_end_5_axes_0"), val = tensor([-1])]; + tensor to_end_5_cast_fp16 = squeeze(axes = to_end_5_axes_0, x = var_1693_cast_fp16)[name = string("to_end_5_cast_fp16")]; + bool var_1695_transpose_x_0 = const()[name = string("op_1695_transpose_x_0"), val = bool(false)]; + bool var_1695_transpose_y_0 = const()[name = string("op_1695_transpose_y_0"), val = bool(false)]; + tensor var_1695_cast_fp16 = matmul(transpose_x = var_1695_transpose_x_0, transpose_y = var_1695_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_5_cast_fp16)[name = string("op_1695_cast_fp16")]; + tensor var_1696 = const()[name = string("op_1696"), val = tensor([32, 16, 64, 64])]; + tensor pair_29_cast_fp16 = reshape(shape = var_1696, x = var_1695_cast_fp16)[name = string("pair_29_cast_fp16")]; + tensor var_1698_cast_fp16 = exp(x = pair_29_cast_fp16)[name = string("op_1698_cast_fp16")]; + tensor pair_decay_5_cast_fp16 = mul(x = var_1698_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_5_cast_fp16")]; + tensor k_beta_5_cast_fp16 = mul(x = k_23_cast_fp16, y = beta_11_cast_fp16)[name = string("k_beta_5_cast_fp16")]; + tensor v_beta_5_cast_fp16 = mul(x = v_11_cast_fp16, y = beta_11_cast_fp16)[name = string("v_beta_5_cast_fp16")]; + bool var_1703_transpose_x_1 = const()[name = string("op_1703_transpose_x_1"), val = bool(false)]; + bool var_1703_transpose_y_1 = const()[name = string("op_1703_transpose_y_1"), val = bool(true)]; + tensor var_1703_cast_fp16 = matmul(transpose_x = var_1703_transpose_x_1, transpose_y = var_1703_transpose_y_1, x = k_beta_5_cast_fp16, y = k_23_cast_fp16)[name = string("op_1703_cast_fp16")]; + tensor kkt_5_cast_fp16 = mul(x = var_1703_cast_fp16, y = pair_decay_5_cast_fp16)[name = string("kkt_5_cast_fp16")]; + bool var_1706_transpose_x_1 = const()[name = string("op_1706_transpose_x_1"), val = bool(false)]; + bool var_1706_transpose_y_1 = const()[name = string("op_1706_transpose_y_1"), val = bool(true)]; + tensor var_1706_cast_fp16 = matmul(transpose_x = var_1706_transpose_x_1, transpose_y = var_1706_transpose_y_1, x = q_29_cast_fp16, y = k_23_cast_fp16)[name = string("op_1706_cast_fp16")]; + tensor attn_intra_5_cast_fp16 = mul(x = var_1706_cast_fp16, y = pair_decay_5_cast_fp16)[name = string("attn_intra_5_cast_fp16")]; + tensor var_1708_cast_fp16 = mul(x = kkt_5_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_1708_cast_fp16")]; + tensor t_9_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_1708_cast_fp16)[name = string("t_9_cast_fp16")]; + tensor var_1712 = const()[name = string("op_1712"), val = tensor([-1, 64, 64])]; + tensor t_11_cast_fp16 = reshape(shape = var_1712, x = t_9_cast_fp16)[name = string("t_11_cast_fp16")]; + tensor var_1717 = const()[name = string("op_1717"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_1718_cast_fp16 = reshape(shape = var_1717, x = t_11_cast_fp16)[name = string("op_1718_cast_fp16")]; + tensor var_1723_cast_fp16 = mul(x = var_1718_cast_fp16, y = var_216_to_fp16)[name = string("op_1723_cast_fp16")]; + tensor blocks_25_axes_0 = const()[name = string("blocks_25_axes_0"), val = tensor([-2])]; + bool blocks_25_keep_dims_0 = const()[name = string("blocks_25_keep_dims_0"), val = bool(false)]; + tensor blocks_25_cast_fp16 = reduce_sum(axes = blocks_25_axes_0, keep_dims = blocks_25_keep_dims_0, x = var_1723_cast_fp16)[name = string("blocks_25_cast_fp16")]; + tensor var_1726_begin_0 = const()[name = string("op_1726_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1726_end_0 = const()[name = string("op_1726_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_1726_end_mask_0 = const()[name = string("op_1726_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1726_cast_fp16 = slice_by_index(begin = var_1726_begin_0, end = var_1726_end_0, end_mask = var_1726_end_mask_0, x = blocks_25_cast_fp16)[name = string("op_1726_cast_fp16")]; + tensor x_57_begin_0 = const()[name = string("x_57_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_57_end_0 = const()[name = string("x_57_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_57_end_mask_0 = const()[name = string("x_57_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_57_cast_fp16 = slice_by_index(begin = x_57_begin_0, end = x_57_end_0, end_mask = x_57_end_mask_0, x = var_1726_cast_fp16)[name = string("x_57_cast_fp16")]; + bool var_1736_transpose_x_0 = const()[name = string("op_1736_transpose_x_0"), val = bool(false)]; + bool var_1736_transpose_y_0 = const()[name = string("op_1736_transpose_y_0"), val = bool(false)]; + tensor var_1736_cast_fp16 = matmul(transpose_x = var_1736_transpose_x_0, transpose_y = var_1736_transpose_y_0, x = x_57_cast_fp16, y = var_224_to_fp16)[name = string("op_1736_cast_fp16")]; + bool var_1737_transpose_x_0 = const()[name = string("op_1737_transpose_x_0"), val = bool(false)]; + bool var_1737_transpose_y_0 = const()[name = string("op_1737_transpose_y_0"), val = bool(false)]; + tensor var_1737_cast_fp16 = matmul(transpose_x = var_1737_transpose_x_0, transpose_y = var_1737_transpose_y_0, x = var_224_to_fp16, y = var_1736_cast_fp16)[name = string("op_1737_cast_fp16")]; + fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_25_cast_fp16 = mul(x = var_1737_cast_fp16, y = const_30_promoted_to_fp16)[name = string("lower_25_cast_fp16")]; + bool bottom_25_interleave_0 = const()[name = string("bottom_25_interleave_0"), val = bool(false)]; + tensor bottom_25_cast_fp16 = concat(axis = var_1564, interleave = bottom_25_interleave_0, values = (lower_25_cast_fp16, var_224_to_fp16))[name = string("bottom_25_cast_fp16")]; + bool inv_35_interleave_0 = const()[name = string("inv_35_interleave_0"), val = bool(false)]; + tensor inv_35_cast_fp16 = concat(axis = var_1543, interleave = inv_35_interleave_0, values = (top_1_to_fp16, bottom_25_cast_fp16))[name = string("inv_35_cast_fp16")]; + tensor var_1746 = const()[name = string("op_1746"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_1747_cast_fp16 = reshape(shape = var_1746, x = t_11_cast_fp16)[name = string("op_1747_cast_fp16")]; + tensor var_1752_cast_fp16 = mul(x = var_1747_cast_fp16, y = var_245_to_fp16)[name = string("op_1752_cast_fp16")]; + tensor blocks_27_axes_0 = const()[name = string("blocks_27_axes_0"), val = tensor([-2])]; + bool blocks_27_keep_dims_0 = const()[name = string("blocks_27_keep_dims_0"), val = bool(false)]; + tensor blocks_27_cast_fp16 = reduce_sum(axes = blocks_27_axes_0, keep_dims = blocks_27_keep_dims_0, x = var_1752_cast_fp16)[name = string("blocks_27_cast_fp16")]; + tensor var_1755_begin_0 = const()[name = string("op_1755_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_1755_end_0 = const()[name = string("op_1755_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_1755_end_mask_0 = const()[name = string("op_1755_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1755_cast_fp16 = slice_by_index(begin = var_1755_begin_0, end = var_1755_end_0, end_mask = var_1755_end_mask_0, x = blocks_27_cast_fp16)[name = string("op_1755_cast_fp16")]; + tensor x_59_begin_0 = const()[name = string("x_59_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_59_end_0 = const()[name = string("x_59_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_59_end_mask_0 = const()[name = string("x_59_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_59_cast_fp16 = slice_by_index(begin = x_59_begin_0, end = x_59_end_0, end_mask = x_59_end_mask_0, x = var_1755_cast_fp16)[name = string("x_59_cast_fp16")]; + tensor var_1757 = const()[name = string("op_1757"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_33_cast_fp16 = reshape(shape = var_1757, x = inv_35_cast_fp16)[name = string("pair_33_cast_fp16")]; + tensor var_1759_begin_0 = const()[name = string("op_1759_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1759_end_0 = const()[name = string("op_1759_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_1759_end_mask_0 = const()[name = string("op_1759_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1759_squeeze_mask_0 = const()[name = string("op_1759_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1759_cast_fp16 = slice_by_index(begin = var_1759_begin_0, end = var_1759_end_0, end_mask = var_1759_end_mask_0, squeeze_mask = var_1759_squeeze_mask_0, x = pair_33_cast_fp16)[name = string("op_1759_cast_fp16")]; + tensor var_1762_begin_0 = const()[name = string("op_1762_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1762_end_0 = const()[name = string("op_1762_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_1762_end_mask_0 = const()[name = string("op_1762_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1762_squeeze_mask_0 = const()[name = string("op_1762_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1762_cast_fp16 = slice_by_index(begin = var_1762_begin_0, end = var_1762_end_0, end_mask = var_1762_end_mask_0, squeeze_mask = var_1762_squeeze_mask_0, x = pair_33_cast_fp16)[name = string("op_1762_cast_fp16")]; + bool var_1765_transpose_x_0 = const()[name = string("op_1765_transpose_x_0"), val = bool(false)]; + bool var_1765_transpose_y_0 = const()[name = string("op_1765_transpose_y_0"), val = bool(false)]; + tensor var_1765_cast_fp16 = matmul(transpose_x = var_1765_transpose_x_0, transpose_y = var_1765_transpose_y_0, x = x_59_cast_fp16, y = var_1759_cast_fp16)[name = string("op_1765_cast_fp16")]; + bool var_1766_transpose_x_0 = const()[name = string("op_1766_transpose_x_0"), val = bool(false)]; + bool var_1766_transpose_y_0 = const()[name = string("op_1766_transpose_y_0"), val = bool(false)]; + tensor var_1766_cast_fp16 = matmul(transpose_x = var_1766_transpose_x_0, transpose_y = var_1766_transpose_y_0, x = var_1762_cast_fp16, y = var_1765_cast_fp16)[name = string("op_1766_cast_fp16")]; + fp16 const_31_promoted_to_fp16 = const()[name = string("const_31_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_27_cast_fp16 = mul(x = var_1766_cast_fp16, y = const_31_promoted_to_fp16)[name = string("lower_27_cast_fp16")]; + bool top_27_interleave_0 = const()[name = string("top_27_interleave_0"), val = bool(false)]; + tensor top_27_cast_fp16 = concat(axis = var_1564, interleave = top_27_interleave_0, values = (var_1759_cast_fp16, var_262_to_fp16))[name = string("top_27_cast_fp16")]; + bool bottom_27_interleave_0 = const()[name = string("bottom_27_interleave_0"), val = bool(false)]; + tensor bottom_27_cast_fp16 = concat(axis = var_1564, interleave = bottom_27_interleave_0, values = (lower_27_cast_fp16, var_1762_cast_fp16))[name = string("bottom_27_cast_fp16")]; + bool inv_37_interleave_0 = const()[name = string("inv_37_interleave_0"), val = bool(false)]; + tensor inv_37_cast_fp16 = concat(axis = var_1543, interleave = inv_37_interleave_0, values = (top_27_cast_fp16, bottom_27_cast_fp16))[name = string("inv_37_cast_fp16")]; + tensor var_1775 = const()[name = string("op_1775"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_1776_cast_fp16 = reshape(shape = var_1775, x = t_11_cast_fp16)[name = string("op_1776_cast_fp16")]; + tensor var_1780_to_fp16 = const()[name = string("op_1780_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518799232)))]; + tensor var_1781_cast_fp16 = mul(x = var_1776_cast_fp16, y = var_1780_to_fp16)[name = string("op_1781_cast_fp16")]; + tensor blocks_29_axes_0 = const()[name = string("blocks_29_axes_0"), val = tensor([-2])]; + bool blocks_29_keep_dims_0 = const()[name = string("blocks_29_keep_dims_0"), val = bool(false)]; + tensor blocks_29_cast_fp16 = reduce_sum(axes = blocks_29_axes_0, keep_dims = blocks_29_keep_dims_0, x = var_1781_cast_fp16)[name = string("blocks_29_cast_fp16")]; + tensor var_1784_begin_0 = const()[name = string("op_1784_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_1784_end_0 = const()[name = string("op_1784_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_1784_end_mask_0 = const()[name = string("op_1784_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1784_cast_fp16 = slice_by_index(begin = var_1784_begin_0, end = var_1784_end_0, end_mask = var_1784_end_mask_0, x = blocks_29_cast_fp16)[name = string("op_1784_cast_fp16")]; + tensor x_61_begin_0 = const()[name = string("x_61_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_61_end_0 = const()[name = string("x_61_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_61_end_mask_0 = const()[name = string("x_61_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_61_cast_fp16 = slice_by_index(begin = x_61_begin_0, end = x_61_end_0, end_mask = x_61_end_mask_0, x = var_1784_cast_fp16)[name = string("x_61_cast_fp16")]; + tensor var_1786 = const()[name = string("op_1786"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_35_cast_fp16 = reshape(shape = var_1786, x = inv_37_cast_fp16)[name = string("pair_35_cast_fp16")]; + tensor var_1788_begin_0 = const()[name = string("op_1788_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1788_end_0 = const()[name = string("op_1788_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_1788_end_mask_0 = const()[name = string("op_1788_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1788_squeeze_mask_0 = const()[name = string("op_1788_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1788_cast_fp16 = slice_by_index(begin = var_1788_begin_0, end = var_1788_end_0, end_mask = var_1788_end_mask_0, squeeze_mask = var_1788_squeeze_mask_0, x = pair_35_cast_fp16)[name = string("op_1788_cast_fp16")]; + tensor var_1791_begin_0 = const()[name = string("op_1791_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1791_end_0 = const()[name = string("op_1791_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_1791_end_mask_0 = const()[name = string("op_1791_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1791_squeeze_mask_0 = const()[name = string("op_1791_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, squeeze_mask = var_1791_squeeze_mask_0, x = pair_35_cast_fp16)[name = string("op_1791_cast_fp16")]; + bool var_1794_transpose_x_0 = const()[name = string("op_1794_transpose_x_0"), val = bool(false)]; + bool var_1794_transpose_y_0 = const()[name = string("op_1794_transpose_y_0"), val = bool(false)]; + tensor var_1794_cast_fp16 = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = x_61_cast_fp16, y = var_1788_cast_fp16)[name = string("op_1794_cast_fp16")]; + bool var_1795_transpose_x_0 = const()[name = string("op_1795_transpose_x_0"), val = bool(false)]; + bool var_1795_transpose_y_0 = const()[name = string("op_1795_transpose_y_0"), val = bool(false)]; + tensor var_1795_cast_fp16 = matmul(transpose_x = var_1795_transpose_x_0, transpose_y = var_1795_transpose_y_0, x = var_1791_cast_fp16, y = var_1794_cast_fp16)[name = string("op_1795_cast_fp16")]; + fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_29_cast_fp16 = mul(x = var_1795_cast_fp16, y = const_32_promoted_to_fp16)[name = string("lower_29_cast_fp16")]; + bool top_29_interleave_0 = const()[name = string("top_29_interleave_0"), val = bool(false)]; + tensor top_29_cast_fp16 = concat(axis = var_1564, interleave = top_29_interleave_0, values = (var_1788_cast_fp16, var_291_to_fp16))[name = string("top_29_cast_fp16")]; + bool bottom_29_interleave_0 = const()[name = string("bottom_29_interleave_0"), val = bool(false)]; + tensor bottom_29_cast_fp16 = concat(axis = var_1564, interleave = bottom_29_interleave_0, values = (lower_29_cast_fp16, var_1791_cast_fp16))[name = string("bottom_29_cast_fp16")]; + bool inv_39_interleave_0 = const()[name = string("inv_39_interleave_0"), val = bool(false)]; + tensor inv_39_cast_fp16 = concat(axis = var_1543, interleave = inv_39_interleave_0, values = (top_29_cast_fp16, bottom_29_cast_fp16))[name = string("inv_39_cast_fp16")]; + tensor var_1804 = const()[name = string("op_1804"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_1805_cast_fp16 = reshape(shape = var_1804, x = t_11_cast_fp16)[name = string("op_1805_cast_fp16")]; + tensor var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518799424)))]; + tensor var_1810_cast_fp16 = mul(x = var_1805_cast_fp16, y = var_1809_to_fp16)[name = string("op_1810_cast_fp16")]; + tensor blocks_31_axes_0 = const()[name = string("blocks_31_axes_0"), val = tensor([-2])]; + bool blocks_31_keep_dims_0 = const()[name = string("blocks_31_keep_dims_0"), val = bool(false)]; + tensor blocks_31_cast_fp16 = reduce_sum(axes = blocks_31_axes_0, keep_dims = blocks_31_keep_dims_0, x = var_1810_cast_fp16)[name = string("blocks_31_cast_fp16")]; + tensor var_1813_begin_0 = const()[name = string("op_1813_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_1813_end_0 = const()[name = string("op_1813_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_1813_end_mask_0 = const()[name = string("op_1813_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1813_cast_fp16 = slice_by_index(begin = var_1813_begin_0, end = var_1813_end_0, end_mask = var_1813_end_mask_0, x = blocks_31_cast_fp16)[name = string("op_1813_cast_fp16")]; + tensor x_63_begin_0 = const()[name = string("x_63_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_63_end_0 = const()[name = string("x_63_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_63_end_mask_0 = const()[name = string("x_63_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_63_cast_fp16 = slice_by_index(begin = x_63_begin_0, end = x_63_end_0, end_mask = x_63_end_mask_0, x = var_1813_cast_fp16)[name = string("x_63_cast_fp16")]; + tensor var_1815 = const()[name = string("op_1815"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_37_cast_fp16 = reshape(shape = var_1815, x = inv_39_cast_fp16)[name = string("pair_37_cast_fp16")]; + tensor var_1817_begin_0 = const()[name = string("op_1817_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1817_end_0 = const()[name = string("op_1817_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_1817_end_mask_0 = const()[name = string("op_1817_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1817_squeeze_mask_0 = const()[name = string("op_1817_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1817_cast_fp16 = slice_by_index(begin = var_1817_begin_0, end = var_1817_end_0, end_mask = var_1817_end_mask_0, squeeze_mask = var_1817_squeeze_mask_0, x = pair_37_cast_fp16)[name = string("op_1817_cast_fp16")]; + tensor var_1820_begin_0 = const()[name = string("op_1820_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1820_end_0 = const()[name = string("op_1820_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_1820_end_mask_0 = const()[name = string("op_1820_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1820_squeeze_mask_0 = const()[name = string("op_1820_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, squeeze_mask = var_1820_squeeze_mask_0, x = pair_37_cast_fp16)[name = string("op_1820_cast_fp16")]; + bool var_1823_transpose_x_0 = const()[name = string("op_1823_transpose_x_0"), val = bool(false)]; + bool var_1823_transpose_y_0 = const()[name = string("op_1823_transpose_y_0"), val = bool(false)]; + tensor var_1823_cast_fp16 = matmul(transpose_x = var_1823_transpose_x_0, transpose_y = var_1823_transpose_y_0, x = x_63_cast_fp16, y = var_1817_cast_fp16)[name = string("op_1823_cast_fp16")]; + bool var_1824_transpose_x_0 = const()[name = string("op_1824_transpose_x_0"), val = bool(false)]; + bool var_1824_transpose_y_0 = const()[name = string("op_1824_transpose_y_0"), val = bool(false)]; + tensor var_1824_cast_fp16 = matmul(transpose_x = var_1824_transpose_x_0, transpose_y = var_1824_transpose_y_0, x = var_1820_cast_fp16, y = var_1823_cast_fp16)[name = string("op_1824_cast_fp16")]; + fp16 const_33_promoted_to_fp16 = const()[name = string("const_33_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_31_cast_fp16 = mul(x = var_1824_cast_fp16, y = const_33_promoted_to_fp16)[name = string("lower_31_cast_fp16")]; + bool top_31_interleave_0 = const()[name = string("top_31_interleave_0"), val = bool(false)]; + tensor top_31_cast_fp16 = concat(axis = var_1564, interleave = top_31_interleave_0, values = (var_1817_cast_fp16, var_320_to_fp16))[name = string("top_31_cast_fp16")]; + bool bottom_31_interleave_0 = const()[name = string("bottom_31_interleave_0"), val = bool(false)]; + tensor bottom_31_cast_fp16 = concat(axis = var_1564, interleave = bottom_31_interleave_0, values = (lower_31_cast_fp16, var_1820_cast_fp16))[name = string("bottom_31_cast_fp16")]; + bool inv_41_interleave_0 = const()[name = string("inv_41_interleave_0"), val = bool(false)]; + tensor inv_41_cast_fp16 = concat(axis = var_1543, interleave = inv_41_interleave_0, values = (top_31_cast_fp16, bottom_31_cast_fp16))[name = string("inv_41_cast_fp16")]; + tensor var_1833 = const()[name = string("op_1833"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_1834_cast_fp16 = reshape(shape = var_1833, x = t_11_cast_fp16)[name = string("op_1834_cast_fp16")]; + tensor var_1838_to_fp16 = const()[name = string("op_1838_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_1839_cast_fp16 = mul(x = var_1834_cast_fp16, y = var_1838_to_fp16)[name = string("op_1839_cast_fp16")]; + tensor blocks_33_axes_0 = const()[name = string("blocks_33_axes_0"), val = tensor([-2])]; + bool blocks_33_keep_dims_0 = const()[name = string("blocks_33_keep_dims_0"), val = bool(false)]; + tensor blocks_33_cast_fp16 = reduce_sum(axes = blocks_33_axes_0, keep_dims = blocks_33_keep_dims_0, x = var_1839_cast_fp16)[name = string("blocks_33_cast_fp16")]; + tensor var_1842_begin_0 = const()[name = string("op_1842_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_1842_end_0 = const()[name = string("op_1842_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_1842_end_mask_0 = const()[name = string("op_1842_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1842_cast_fp16 = slice_by_index(begin = var_1842_begin_0, end = var_1842_end_0, end_mask = var_1842_end_mask_0, x = blocks_33_cast_fp16)[name = string("op_1842_cast_fp16")]; + tensor x_65_begin_0 = const()[name = string("x_65_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_65_end_0 = const()[name = string("x_65_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_65_end_mask_0 = const()[name = string("x_65_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_65_cast_fp16 = slice_by_index(begin = x_65_begin_0, end = x_65_end_0, end_mask = x_65_end_mask_0, x = var_1842_cast_fp16)[name = string("x_65_cast_fp16")]; + tensor var_1844 = const()[name = string("op_1844"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_39_cast_fp16 = reshape(shape = var_1844, x = inv_41_cast_fp16)[name = string("pair_39_cast_fp16")]; + tensor var_1846_begin_0 = const()[name = string("op_1846_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1846_end_0 = const()[name = string("op_1846_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_1846_end_mask_0 = const()[name = string("op_1846_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1846_squeeze_mask_0 = const()[name = string("op_1846_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1846_cast_fp16 = slice_by_index(begin = var_1846_begin_0, end = var_1846_end_0, end_mask = var_1846_end_mask_0, squeeze_mask = var_1846_squeeze_mask_0, x = pair_39_cast_fp16)[name = string("op_1846_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1849_squeeze_mask_0 = const()[name = string("op_1849_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, squeeze_mask = var_1849_squeeze_mask_0, x = pair_39_cast_fp16)[name = string("op_1849_cast_fp16")]; + bool var_1852_transpose_x_0 = const()[name = string("op_1852_transpose_x_0"), val = bool(false)]; + bool var_1852_transpose_y_0 = const()[name = string("op_1852_transpose_y_0"), val = bool(false)]; + tensor var_1852_cast_fp16 = matmul(transpose_x = var_1852_transpose_x_0, transpose_y = var_1852_transpose_y_0, x = x_65_cast_fp16, y = var_1846_cast_fp16)[name = string("op_1852_cast_fp16")]; + bool var_1853_transpose_x_0 = const()[name = string("op_1853_transpose_x_0"), val = bool(false)]; + bool var_1853_transpose_y_0 = const()[name = string("op_1853_transpose_y_0"), val = bool(false)]; + tensor var_1853_cast_fp16 = matmul(transpose_x = var_1853_transpose_x_0, transpose_y = var_1853_transpose_y_0, x = var_1849_cast_fp16, y = var_1852_cast_fp16)[name = string("op_1853_cast_fp16")]; + fp16 const_34_promoted_to_fp16 = const()[name = string("const_34_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_33_cast_fp16 = mul(x = var_1853_cast_fp16, y = const_34_promoted_to_fp16)[name = string("lower_33_cast_fp16")]; + bool top_33_interleave_0 = const()[name = string("top_33_interleave_0"), val = bool(false)]; + tensor top_33_cast_fp16 = concat(axis = var_1564, interleave = top_33_interleave_0, values = (var_1846_cast_fp16, var_349_to_fp16))[name = string("top_33_cast_fp16")]; + bool bottom_33_interleave_0 = const()[name = string("bottom_33_interleave_0"), val = bool(false)]; + tensor bottom_33_cast_fp16 = concat(axis = var_1564, interleave = bottom_33_interleave_0, values = (lower_33_cast_fp16, var_1849_cast_fp16))[name = string("bottom_33_cast_fp16")]; + bool inv_43_interleave_0 = const()[name = string("inv_43_interleave_0"), val = bool(false)]; + tensor inv_43_cast_fp16 = concat(axis = var_1543, interleave = inv_43_interleave_0, values = (top_33_cast_fp16, bottom_33_cast_fp16))[name = string("inv_43_cast_fp16")]; + tensor var_1862 = const()[name = string("op_1862"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_1863_cast_fp16 = reshape(shape = var_1862, x = t_11_cast_fp16)[name = string("op_1863_cast_fp16")]; + tensor blocks_35_axes_0 = const()[name = string("blocks_35_axes_0"), val = tensor([-2])]; + bool blocks_35_keep_dims_0 = const()[name = string("blocks_35_keep_dims_0"), val = bool(false)]; + tensor blocks_35_cast_fp16 = reduce_sum(axes = blocks_35_axes_0, keep_dims = blocks_35_keep_dims_0, x = var_1863_cast_fp16)[name = string("blocks_35_cast_fp16")]; + tensor var_1871_begin_0 = const()[name = string("op_1871_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_1871_end_0 = const()[name = string("op_1871_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_1871_end_mask_0 = const()[name = string("op_1871_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1871_cast_fp16 = slice_by_index(begin = var_1871_begin_0, end = var_1871_end_0, end_mask = var_1871_end_mask_0, x = blocks_35_cast_fp16)[name = string("op_1871_cast_fp16")]; + tensor x_67_begin_0 = const()[name = string("x_67_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_67_end_0 = const()[name = string("x_67_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_67_end_mask_0 = const()[name = string("x_67_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_67_cast_fp16 = slice_by_index(begin = x_67_begin_0, end = x_67_end_0, end_mask = x_67_end_mask_0, x = var_1871_cast_fp16)[name = string("x_67_cast_fp16")]; + tensor var_1873 = const()[name = string("op_1873"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_41_cast_fp16 = reshape(shape = var_1873, x = inv_43_cast_fp16)[name = string("pair_41_cast_fp16")]; + tensor var_1875_begin_0 = const()[name = string("op_1875_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_1875_end_0 = const()[name = string("op_1875_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_1875_end_mask_0 = const()[name = string("op_1875_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1875_squeeze_mask_0 = const()[name = string("op_1875_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1875_cast_fp16 = slice_by_index(begin = var_1875_begin_0, end = var_1875_end_0, end_mask = var_1875_end_mask_0, squeeze_mask = var_1875_squeeze_mask_0, x = pair_41_cast_fp16)[name = string("op_1875_cast_fp16")]; + tensor var_1878_begin_0 = const()[name = string("op_1878_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_1878_end_0 = const()[name = string("op_1878_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_1878_end_mask_0 = const()[name = string("op_1878_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_1878_squeeze_mask_0 = const()[name = string("op_1878_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_1878_cast_fp16 = slice_by_index(begin = var_1878_begin_0, end = var_1878_end_0, end_mask = var_1878_end_mask_0, squeeze_mask = var_1878_squeeze_mask_0, x = pair_41_cast_fp16)[name = string("op_1878_cast_fp16")]; + bool var_1881_transpose_x_0 = const()[name = string("op_1881_transpose_x_0"), val = bool(false)]; + bool var_1881_transpose_y_0 = const()[name = string("op_1881_transpose_y_0"), val = bool(false)]; + tensor var_1881_cast_fp16 = matmul(transpose_x = var_1881_transpose_x_0, transpose_y = var_1881_transpose_y_0, x = x_67_cast_fp16, y = var_1875_cast_fp16)[name = string("op_1881_cast_fp16")]; + bool var_1882_transpose_x_0 = const()[name = string("op_1882_transpose_x_0"), val = bool(false)]; + bool var_1882_transpose_y_0 = const()[name = string("op_1882_transpose_y_0"), val = bool(false)]; + tensor var_1882_cast_fp16 = matmul(transpose_x = var_1882_transpose_x_0, transpose_y = var_1882_transpose_y_0, x = var_1878_cast_fp16, y = var_1881_cast_fp16)[name = string("op_1882_cast_fp16")]; + fp16 const_35_promoted_to_fp16 = const()[name = string("const_35_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_35_cast_fp16 = mul(x = var_1882_cast_fp16, y = const_35_promoted_to_fp16)[name = string("lower_35_cast_fp16")]; + bool top_35_interleave_0 = const()[name = string("top_35_interleave_0"), val = bool(false)]; + tensor top_35_cast_fp16 = concat(axis = var_1564, interleave = top_35_interleave_0, values = (var_1875_cast_fp16, var_378_to_fp16))[name = string("top_35_cast_fp16")]; + bool bottom_35_interleave_0 = const()[name = string("bottom_35_interleave_0"), val = bool(false)]; + tensor bottom_35_cast_fp16 = concat(axis = var_1564, interleave = bottom_35_interleave_0, values = (lower_35_cast_fp16, var_1878_cast_fp16))[name = string("bottom_35_cast_fp16")]; + bool inv_45_interleave_0 = const()[name = string("inv_45_interleave_0"), val = bool(false)]; + tensor inv_45_cast_fp16 = concat(axis = var_1543, interleave = inv_45_interleave_0, values = (top_35_cast_fp16, bottom_35_cast_fp16))[name = string("inv_45_cast_fp16")]; + tensor var_1891 = const()[name = string("op_1891"), val = tensor([32, 16, 64, 64])]; + tensor inv_47_cast_fp16 = reshape(shape = var_1891, x = inv_45_cast_fp16)[name = string("inv_47_cast_fp16")]; + bool new_v_5_transpose_x_0 = const()[name = string("new_v_5_transpose_x_0"), val = bool(false)]; + bool new_v_5_transpose_y_0 = const()[name = string("new_v_5_transpose_y_0"), val = bool(false)]; + tensor new_v_5_cast_fp16 = matmul(transpose_x = new_v_5_transpose_x_0, transpose_y = new_v_5_transpose_y_0, x = inv_47_cast_fp16, y = v_beta_5_cast_fp16)[name = string("new_v_5_cast_fp16")]; + tensor var_1894_cast_fp16 = exp(x = cum_5_cast_fp16)[name = string("op_1894_cast_fp16")]; + tensor var_1895_axes_0 = const()[name = string("op_1895_axes_0"), val = tensor([3])]; + tensor var_1895_cast_fp16 = expand_dims(axes = var_1895_axes_0, x = var_1894_cast_fp16)[name = string("op_1895_cast_fp16")]; + tensor var_1896_cast_fp16 = mul(x = k_beta_5_cast_fp16, y = var_1895_cast_fp16)[name = string("op_1896_cast_fp16")]; + bool k_cumdecay_5_transpose_x_0 = const()[name = string("k_cumdecay_5_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_5_transpose_y_0 = const()[name = string("k_cumdecay_5_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_5_cast_fp16 = matmul(transpose_x = k_cumdecay_5_transpose_x_0, transpose_y = k_cumdecay_5_transpose_y_0, x = inv_47_cast_fp16, y = var_1896_cast_fp16)[name = string("k_cumdecay_5_cast_fp16")]; + tensor q_dec_5_cast_fp16 = mul(x = q_29_cast_fp16, y = var_1895_cast_fp16)[name = string("q_dec_5_cast_fp16")]; + tensor var_1901_cast_fp16 = exp(x = to_end_5_cast_fp16)[name = string("op_1901_cast_fp16")]; + tensor var_1902_axes_0 = const()[name = string("op_1902_axes_0"), val = tensor([3])]; + tensor var_1902_cast_fp16 = expand_dims(axes = var_1902_axes_0, x = var_1901_cast_fp16)[name = string("op_1902_cast_fp16")]; + tensor k_dec_5_cast_fp16 = mul(x = k_23_cast_fp16, y = var_1902_cast_fp16)[name = string("k_dec_5_cast_fp16")]; + tensor var_1904_begin_0 = const()[name = string("op_1904_begin_0"), val = tensor([0, 0, -1])]; + tensor var_1904_end_0 = const()[name = string("op_1904_end_0"), val = tensor([32, 16, 64])]; + tensor var_1904_end_mask_0 = const()[name = string("op_1904_end_mask_0"), val = tensor([true, true, true])]; + tensor var_1904_cast_fp16 = slice_by_index(begin = var_1904_begin_0, end = var_1904_end_0, end_mask = var_1904_end_mask_0, x = cum_5_cast_fp16)[name = string("op_1904_cast_fp16")]; + tensor var_1905_cast_fp16 = exp(x = var_1904_cast_fp16)[name = string("op_1905_cast_fp16")]; + tensor chunk_decay_5_axes_0 = const()[name = string("chunk_decay_5_axes_0"), val = tensor([3])]; + tensor chunk_decay_5_cast_fp16 = expand_dims(axes = chunk_decay_5_axes_0, x = var_1905_cast_fp16)[name = string("chunk_decay_5_cast_fp16")]; + tensor var_1910_begin_0 = const()[name = string("op_1910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1910_end_0 = const()[name = string("op_1910_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1910_end_mask_0 = const()[name = string("op_1910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1910_squeeze_mask_0 = const()[name = string("op_1910_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1910_cast_fp16 = slice_by_index(begin = var_1910_begin_0, end = var_1910_end_0, end_mask = var_1910_end_mask_0, squeeze_mask = var_1910_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_1910_cast_fp16")]; + tensor var_1912_begin_0 = const()[name = string("op_1912_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1912_end_0 = const()[name = string("op_1912_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1912_end_mask_0 = const()[name = string("op_1912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1912_squeeze_mask_0 = const()[name = string("op_1912_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1912_cast_fp16 = slice_by_index(begin = var_1912_begin_0, end = var_1912_end_0, end_mask = var_1912_end_mask_0, squeeze_mask = var_1912_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_1912_cast_fp16")]; + bool var_1913_transpose_x_0 = const()[name = string("op_1913_transpose_x_0"), val = bool(false)]; + bool var_1913_transpose_y_0 = const()[name = string("op_1913_transpose_y_0"), val = bool(false)]; + tensor var_1913_cast_fp16 = matmul(transpose_x = var_1913_transpose_x_0, transpose_y = var_1913_transpose_y_0, x = var_1912_cast_fp16, y = state_1_to_fp16)[name = string("op_1913_cast_fp16")]; + tensor v_new_65_cast_fp16 = sub(x = var_1910_cast_fp16, y = var_1913_cast_fp16)[name = string("v_new_65_cast_fp16")]; + tensor var_1916_begin_0 = const()[name = string("op_1916_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1916_end_0 = const()[name = string("op_1916_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1916_end_mask_0 = const()[name = string("op_1916_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1916_squeeze_mask_0 = const()[name = string("op_1916_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1916_cast_fp16 = slice_by_index(begin = var_1916_begin_0, end = var_1916_end_0, end_mask = var_1916_end_mask_0, squeeze_mask = var_1916_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_1916_cast_fp16")]; + bool var_1917_transpose_x_0 = const()[name = string("op_1917_transpose_x_0"), val = bool(false)]; + bool var_1917_transpose_y_0 = const()[name = string("op_1917_transpose_y_0"), val = bool(false)]; + tensor var_1917_cast_fp16 = matmul(transpose_x = var_1917_transpose_x_0, transpose_y = var_1917_transpose_y_0, x = var_1916_cast_fp16, y = state_1_to_fp16)[name = string("op_1917_cast_fp16")]; + tensor var_1919_begin_0 = const()[name = string("op_1919_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1919_end_0 = const()[name = string("op_1919_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_1919_end_mask_0 = const()[name = string("op_1919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1919_squeeze_mask_0 = const()[name = string("op_1919_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1919_cast_fp16 = slice_by_index(begin = var_1919_begin_0, end = var_1919_end_0, end_mask = var_1919_end_mask_0, squeeze_mask = var_1919_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_1919_cast_fp16")]; + bool var_1920_transpose_x_0 = const()[name = string("op_1920_transpose_x_0"), val = bool(false)]; + bool var_1920_transpose_y_0 = const()[name = string("op_1920_transpose_y_0"), val = bool(false)]; + tensor var_1920_cast_fp16 = matmul(transpose_x = var_1920_transpose_x_0, transpose_y = var_1920_transpose_y_0, x = var_1919_cast_fp16, y = v_new_65_cast_fp16)[name = string("op_1920_cast_fp16")]; + tensor var_1921_cast_fp16 = add(x = var_1917_cast_fp16, y = var_1920_cast_fp16)[name = string("op_1921_cast_fp16")]; + tensor var_1923_begin_0 = const()[name = string("op_1923_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1923_end_0 = const()[name = string("op_1923_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_1923_end_mask_0 = const()[name = string("op_1923_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1923_squeeze_mask_0 = const()[name = string("op_1923_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1923_cast_fp16 = slice_by_index(begin = var_1923_begin_0, end = var_1923_end_0, end_mask = var_1923_end_mask_0, squeeze_mask = var_1923_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_1923_cast_fp16")]; + tensor var_1924_cast_fp16 = mul(x = state_1_to_fp16, y = var_1923_cast_fp16)[name = string("op_1924_cast_fp16")]; + tensor var_1926_begin_0 = const()[name = string("op_1926_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1926_end_0 = const()[name = string("op_1926_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_1926_end_mask_0 = const()[name = string("op_1926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1926_squeeze_mask_0 = const()[name = string("op_1926_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1926_cast_fp16 = slice_by_index(begin = var_1926_begin_0, end = var_1926_end_0, end_mask = var_1926_end_mask_0, squeeze_mask = var_1926_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_1926_cast_fp16")]; + bool var_1928_transpose_x_1 = const()[name = string("op_1928_transpose_x_1"), val = bool(true)]; + bool var_1928_transpose_y_1 = const()[name = string("op_1928_transpose_y_1"), val = bool(false)]; + tensor var_1928_cast_fp16 = matmul(transpose_x = var_1928_transpose_x_1, transpose_y = var_1928_transpose_y_1, x = var_1926_cast_fp16, y = v_new_65_cast_fp16)[name = string("op_1928_cast_fp16")]; + tensor state_67_cast_fp16 = add(x = var_1924_cast_fp16, y = var_1928_cast_fp16)[name = string("state_67_cast_fp16")]; + tensor var_1931_begin_0 = const()[name = string("op_1931_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1931_end_0 = const()[name = string("op_1931_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1931_end_mask_0 = const()[name = string("op_1931_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1931_squeeze_mask_0 = const()[name = string("op_1931_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1931_cast_fp16 = slice_by_index(begin = var_1931_begin_0, end = var_1931_end_0, end_mask = var_1931_end_mask_0, squeeze_mask = var_1931_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_1931_cast_fp16")]; + tensor var_1933_begin_0 = const()[name = string("op_1933_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1933_end_0 = const()[name = string("op_1933_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1933_end_mask_0 = const()[name = string("op_1933_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1933_squeeze_mask_0 = const()[name = string("op_1933_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1933_cast_fp16 = slice_by_index(begin = var_1933_begin_0, end = var_1933_end_0, end_mask = var_1933_end_mask_0, squeeze_mask = var_1933_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_1933_cast_fp16")]; + bool var_1934_transpose_x_0 = const()[name = string("op_1934_transpose_x_0"), val = bool(false)]; + bool var_1934_transpose_y_0 = const()[name = string("op_1934_transpose_y_0"), val = bool(false)]; + tensor var_1934_cast_fp16 = matmul(transpose_x = var_1934_transpose_x_0, transpose_y = var_1934_transpose_y_0, x = var_1933_cast_fp16, y = state_67_cast_fp16)[name = string("op_1934_cast_fp16")]; + tensor v_new_67_cast_fp16 = sub(x = var_1931_cast_fp16, y = var_1934_cast_fp16)[name = string("v_new_67_cast_fp16")]; + tensor var_1937_begin_0 = const()[name = string("op_1937_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1937_end_0 = const()[name = string("op_1937_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1937_end_mask_0 = const()[name = string("op_1937_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1937_squeeze_mask_0 = const()[name = string("op_1937_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1937_cast_fp16 = slice_by_index(begin = var_1937_begin_0, end = var_1937_end_0, end_mask = var_1937_end_mask_0, squeeze_mask = var_1937_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_1937_cast_fp16")]; + bool var_1938_transpose_x_0 = const()[name = string("op_1938_transpose_x_0"), val = bool(false)]; + bool var_1938_transpose_y_0 = const()[name = string("op_1938_transpose_y_0"), val = bool(false)]; + tensor var_1938_cast_fp16 = matmul(transpose_x = var_1938_transpose_x_0, transpose_y = var_1938_transpose_y_0, x = var_1937_cast_fp16, y = state_67_cast_fp16)[name = string("op_1938_cast_fp16")]; + tensor var_1940_begin_0 = const()[name = string("op_1940_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1940_end_0 = const()[name = string("op_1940_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_1940_end_mask_0 = const()[name = string("op_1940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1940_squeeze_mask_0 = const()[name = string("op_1940_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1940_cast_fp16 = slice_by_index(begin = var_1940_begin_0, end = var_1940_end_0, end_mask = var_1940_end_mask_0, squeeze_mask = var_1940_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_1940_cast_fp16")]; + bool var_1941_transpose_x_0 = const()[name = string("op_1941_transpose_x_0"), val = bool(false)]; + bool var_1941_transpose_y_0 = const()[name = string("op_1941_transpose_y_0"), val = bool(false)]; + tensor var_1941_cast_fp16 = matmul(transpose_x = var_1941_transpose_x_0, transpose_y = var_1941_transpose_y_0, x = var_1940_cast_fp16, y = v_new_67_cast_fp16)[name = string("op_1941_cast_fp16")]; + tensor var_1942_cast_fp16 = add(x = var_1938_cast_fp16, y = var_1941_cast_fp16)[name = string("op_1942_cast_fp16")]; + tensor var_1944_begin_0 = const()[name = string("op_1944_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1944_end_0 = const()[name = string("op_1944_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_1944_end_mask_0 = const()[name = string("op_1944_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1944_squeeze_mask_0 = const()[name = string("op_1944_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1944_cast_fp16 = slice_by_index(begin = var_1944_begin_0, end = var_1944_end_0, end_mask = var_1944_end_mask_0, squeeze_mask = var_1944_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_1944_cast_fp16")]; + tensor var_1945_cast_fp16 = mul(x = state_67_cast_fp16, y = var_1944_cast_fp16)[name = string("op_1945_cast_fp16")]; + tensor var_1947_begin_0 = const()[name = string("op_1947_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_1947_end_0 = const()[name = string("op_1947_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_1947_end_mask_0 = const()[name = string("op_1947_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1947_squeeze_mask_0 = const()[name = string("op_1947_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1947_cast_fp16 = slice_by_index(begin = var_1947_begin_0, end = var_1947_end_0, end_mask = var_1947_end_mask_0, squeeze_mask = var_1947_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_1947_cast_fp16")]; + bool var_1949_transpose_x_1 = const()[name = string("op_1949_transpose_x_1"), val = bool(true)]; + bool var_1949_transpose_y_1 = const()[name = string("op_1949_transpose_y_1"), val = bool(false)]; + tensor var_1949_cast_fp16 = matmul(transpose_x = var_1949_transpose_x_1, transpose_y = var_1949_transpose_y_1, x = var_1947_cast_fp16, y = v_new_67_cast_fp16)[name = string("op_1949_cast_fp16")]; + tensor state_69_cast_fp16 = add(x = var_1945_cast_fp16, y = var_1949_cast_fp16)[name = string("state_69_cast_fp16")]; + tensor var_1952_begin_0 = const()[name = string("op_1952_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1952_end_0 = const()[name = string("op_1952_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1952_end_mask_0 = const()[name = string("op_1952_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1952_squeeze_mask_0 = const()[name = string("op_1952_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1952_cast_fp16 = slice_by_index(begin = var_1952_begin_0, end = var_1952_end_0, end_mask = var_1952_end_mask_0, squeeze_mask = var_1952_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_1952_cast_fp16")]; + tensor var_1954_begin_0 = const()[name = string("op_1954_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1954_end_0 = const()[name = string("op_1954_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1954_end_mask_0 = const()[name = string("op_1954_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1954_squeeze_mask_0 = const()[name = string("op_1954_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1954_cast_fp16 = slice_by_index(begin = var_1954_begin_0, end = var_1954_end_0, end_mask = var_1954_end_mask_0, squeeze_mask = var_1954_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_1954_cast_fp16")]; + bool var_1955_transpose_x_0 = const()[name = string("op_1955_transpose_x_0"), val = bool(false)]; + bool var_1955_transpose_y_0 = const()[name = string("op_1955_transpose_y_0"), val = bool(false)]; + tensor var_1955_cast_fp16 = matmul(transpose_x = var_1955_transpose_x_0, transpose_y = var_1955_transpose_y_0, x = var_1954_cast_fp16, y = state_69_cast_fp16)[name = string("op_1955_cast_fp16")]; + tensor v_new_69_cast_fp16 = sub(x = var_1952_cast_fp16, y = var_1955_cast_fp16)[name = string("v_new_69_cast_fp16")]; + tensor var_1958_begin_0 = const()[name = string("op_1958_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1958_end_0 = const()[name = string("op_1958_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1958_end_mask_0 = const()[name = string("op_1958_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1958_squeeze_mask_0 = const()[name = string("op_1958_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1958_cast_fp16 = slice_by_index(begin = var_1958_begin_0, end = var_1958_end_0, end_mask = var_1958_end_mask_0, squeeze_mask = var_1958_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_1958_cast_fp16")]; + bool var_1959_transpose_x_0 = const()[name = string("op_1959_transpose_x_0"), val = bool(false)]; + bool var_1959_transpose_y_0 = const()[name = string("op_1959_transpose_y_0"), val = bool(false)]; + tensor var_1959_cast_fp16 = matmul(transpose_x = var_1959_transpose_x_0, transpose_y = var_1959_transpose_y_0, x = var_1958_cast_fp16, y = state_69_cast_fp16)[name = string("op_1959_cast_fp16")]; + tensor var_1961_begin_0 = const()[name = string("op_1961_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1961_end_0 = const()[name = string("op_1961_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_1961_end_mask_0 = const()[name = string("op_1961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1961_squeeze_mask_0 = const()[name = string("op_1961_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1961_cast_fp16 = slice_by_index(begin = var_1961_begin_0, end = var_1961_end_0, end_mask = var_1961_end_mask_0, squeeze_mask = var_1961_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_1961_cast_fp16")]; + bool var_1962_transpose_x_0 = const()[name = string("op_1962_transpose_x_0"), val = bool(false)]; + bool var_1962_transpose_y_0 = const()[name = string("op_1962_transpose_y_0"), val = bool(false)]; + tensor var_1962_cast_fp16 = matmul(transpose_x = var_1962_transpose_x_0, transpose_y = var_1962_transpose_y_0, x = var_1961_cast_fp16, y = v_new_69_cast_fp16)[name = string("op_1962_cast_fp16")]; + tensor var_1963_cast_fp16 = add(x = var_1959_cast_fp16, y = var_1962_cast_fp16)[name = string("op_1963_cast_fp16")]; + tensor var_1965_begin_0 = const()[name = string("op_1965_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1965_end_0 = const()[name = string("op_1965_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_1965_end_mask_0 = const()[name = string("op_1965_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1965_squeeze_mask_0 = const()[name = string("op_1965_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1965_cast_fp16 = slice_by_index(begin = var_1965_begin_0, end = var_1965_end_0, end_mask = var_1965_end_mask_0, squeeze_mask = var_1965_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_1965_cast_fp16")]; + tensor var_1966_cast_fp16 = mul(x = state_69_cast_fp16, y = var_1965_cast_fp16)[name = string("op_1966_cast_fp16")]; + tensor var_1968_begin_0 = const()[name = string("op_1968_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_1968_end_0 = const()[name = string("op_1968_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_1968_end_mask_0 = const()[name = string("op_1968_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1968_squeeze_mask_0 = const()[name = string("op_1968_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1968_cast_fp16 = slice_by_index(begin = var_1968_begin_0, end = var_1968_end_0, end_mask = var_1968_end_mask_0, squeeze_mask = var_1968_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_1968_cast_fp16")]; + bool var_1970_transpose_x_1 = const()[name = string("op_1970_transpose_x_1"), val = bool(true)]; + bool var_1970_transpose_y_1 = const()[name = string("op_1970_transpose_y_1"), val = bool(false)]; + tensor var_1970_cast_fp16 = matmul(transpose_x = var_1970_transpose_x_1, transpose_y = var_1970_transpose_y_1, x = var_1968_cast_fp16, y = v_new_69_cast_fp16)[name = string("op_1970_cast_fp16")]; + tensor state_71_cast_fp16 = add(x = var_1966_cast_fp16, y = var_1970_cast_fp16)[name = string("state_71_cast_fp16")]; + tensor var_1973_begin_0 = const()[name = string("op_1973_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1973_end_0 = const()[name = string("op_1973_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1973_end_mask_0 = const()[name = string("op_1973_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1973_squeeze_mask_0 = const()[name = string("op_1973_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1973_cast_fp16 = slice_by_index(begin = var_1973_begin_0, end = var_1973_end_0, end_mask = var_1973_end_mask_0, squeeze_mask = var_1973_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_1973_cast_fp16")]; + tensor var_1975_begin_0 = const()[name = string("op_1975_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1975_end_0 = const()[name = string("op_1975_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1975_end_mask_0 = const()[name = string("op_1975_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1975_squeeze_mask_0 = const()[name = string("op_1975_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1975_cast_fp16 = slice_by_index(begin = var_1975_begin_0, end = var_1975_end_0, end_mask = var_1975_end_mask_0, squeeze_mask = var_1975_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_1975_cast_fp16")]; + bool var_1976_transpose_x_0 = const()[name = string("op_1976_transpose_x_0"), val = bool(false)]; + bool var_1976_transpose_y_0 = const()[name = string("op_1976_transpose_y_0"), val = bool(false)]; + tensor var_1976_cast_fp16 = matmul(transpose_x = var_1976_transpose_x_0, transpose_y = var_1976_transpose_y_0, x = var_1975_cast_fp16, y = state_71_cast_fp16)[name = string("op_1976_cast_fp16")]; + tensor v_new_71_cast_fp16 = sub(x = var_1973_cast_fp16, y = var_1976_cast_fp16)[name = string("v_new_71_cast_fp16")]; + tensor var_1979_begin_0 = const()[name = string("op_1979_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1979_end_0 = const()[name = string("op_1979_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1979_end_mask_0 = const()[name = string("op_1979_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1979_squeeze_mask_0 = const()[name = string("op_1979_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1979_cast_fp16 = slice_by_index(begin = var_1979_begin_0, end = var_1979_end_0, end_mask = var_1979_end_mask_0, squeeze_mask = var_1979_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_1979_cast_fp16")]; + bool var_1980_transpose_x_0 = const()[name = string("op_1980_transpose_x_0"), val = bool(false)]; + bool var_1980_transpose_y_0 = const()[name = string("op_1980_transpose_y_0"), val = bool(false)]; + tensor var_1980_cast_fp16 = matmul(transpose_x = var_1980_transpose_x_0, transpose_y = var_1980_transpose_y_0, x = var_1979_cast_fp16, y = state_71_cast_fp16)[name = string("op_1980_cast_fp16")]; + tensor var_1982_begin_0 = const()[name = string("op_1982_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1982_end_0 = const()[name = string("op_1982_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_1982_end_mask_0 = const()[name = string("op_1982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1982_squeeze_mask_0 = const()[name = string("op_1982_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1982_cast_fp16 = slice_by_index(begin = var_1982_begin_0, end = var_1982_end_0, end_mask = var_1982_end_mask_0, squeeze_mask = var_1982_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_1982_cast_fp16")]; + bool var_1983_transpose_x_0 = const()[name = string("op_1983_transpose_x_0"), val = bool(false)]; + bool var_1983_transpose_y_0 = const()[name = string("op_1983_transpose_y_0"), val = bool(false)]; + tensor var_1983_cast_fp16 = matmul(transpose_x = var_1983_transpose_x_0, transpose_y = var_1983_transpose_y_0, x = var_1982_cast_fp16, y = v_new_71_cast_fp16)[name = string("op_1983_cast_fp16")]; + tensor var_1984_cast_fp16 = add(x = var_1980_cast_fp16, y = var_1983_cast_fp16)[name = string("op_1984_cast_fp16")]; + tensor var_1986_begin_0 = const()[name = string("op_1986_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1986_end_0 = const()[name = string("op_1986_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_1986_end_mask_0 = const()[name = string("op_1986_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1986_squeeze_mask_0 = const()[name = string("op_1986_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1986_cast_fp16 = slice_by_index(begin = var_1986_begin_0, end = var_1986_end_0, end_mask = var_1986_end_mask_0, squeeze_mask = var_1986_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_1986_cast_fp16")]; + tensor var_1987_cast_fp16 = mul(x = state_71_cast_fp16, y = var_1986_cast_fp16)[name = string("op_1987_cast_fp16")]; + tensor var_1989_begin_0 = const()[name = string("op_1989_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_1989_end_0 = const()[name = string("op_1989_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_1989_end_mask_0 = const()[name = string("op_1989_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1989_squeeze_mask_0 = const()[name = string("op_1989_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1989_cast_fp16 = slice_by_index(begin = var_1989_begin_0, end = var_1989_end_0, end_mask = var_1989_end_mask_0, squeeze_mask = var_1989_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_1989_cast_fp16")]; + bool var_1991_transpose_x_1 = const()[name = string("op_1991_transpose_x_1"), val = bool(true)]; + bool var_1991_transpose_y_1 = const()[name = string("op_1991_transpose_y_1"), val = bool(false)]; + tensor var_1991_cast_fp16 = matmul(transpose_x = var_1991_transpose_x_1, transpose_y = var_1991_transpose_y_1, x = var_1989_cast_fp16, y = v_new_71_cast_fp16)[name = string("op_1991_cast_fp16")]; + tensor state_73_cast_fp16 = add(x = var_1987_cast_fp16, y = var_1991_cast_fp16)[name = string("state_73_cast_fp16")]; + tensor var_1994_begin_0 = const()[name = string("op_1994_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1994_end_0 = const()[name = string("op_1994_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1994_end_mask_0 = const()[name = string("op_1994_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1994_squeeze_mask_0 = const()[name = string("op_1994_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1994_cast_fp16 = slice_by_index(begin = var_1994_begin_0, end = var_1994_end_0, end_mask = var_1994_end_mask_0, squeeze_mask = var_1994_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_1994_cast_fp16")]; + tensor var_1996_begin_0 = const()[name = string("op_1996_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_1996_end_0 = const()[name = string("op_1996_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_1996_end_mask_0 = const()[name = string("op_1996_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1996_squeeze_mask_0 = const()[name = string("op_1996_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_1996_cast_fp16 = slice_by_index(begin = var_1996_begin_0, end = var_1996_end_0, end_mask = var_1996_end_mask_0, squeeze_mask = var_1996_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_1996_cast_fp16")]; + bool var_1997_transpose_x_0 = const()[name = string("op_1997_transpose_x_0"), val = bool(false)]; + bool var_1997_transpose_y_0 = const()[name = string("op_1997_transpose_y_0"), val = bool(false)]; + tensor var_1997_cast_fp16 = matmul(transpose_x = var_1997_transpose_x_0, transpose_y = var_1997_transpose_y_0, x = var_1996_cast_fp16, y = state_73_cast_fp16)[name = string("op_1997_cast_fp16")]; + tensor v_new_73_cast_fp16 = sub(x = var_1994_cast_fp16, y = var_1997_cast_fp16)[name = string("v_new_73_cast_fp16")]; + tensor var_2000_begin_0 = const()[name = string("op_2000_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2000_end_0 = const()[name = string("op_2000_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2000_end_mask_0 = const()[name = string("op_2000_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2000_squeeze_mask_0 = const()[name = string("op_2000_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2000_cast_fp16 = slice_by_index(begin = var_2000_begin_0, end = var_2000_end_0, end_mask = var_2000_end_mask_0, squeeze_mask = var_2000_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2000_cast_fp16")]; + bool var_2001_transpose_x_0 = const()[name = string("op_2001_transpose_x_0"), val = bool(false)]; + bool var_2001_transpose_y_0 = const()[name = string("op_2001_transpose_y_0"), val = bool(false)]; + tensor var_2001_cast_fp16 = matmul(transpose_x = var_2001_transpose_x_0, transpose_y = var_2001_transpose_y_0, x = var_2000_cast_fp16, y = state_73_cast_fp16)[name = string("op_2001_cast_fp16")]; + tensor var_2003_begin_0 = const()[name = string("op_2003_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2003_end_0 = const()[name = string("op_2003_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_2003_end_mask_0 = const()[name = string("op_2003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2003_squeeze_mask_0 = const()[name = string("op_2003_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2003_cast_fp16 = slice_by_index(begin = var_2003_begin_0, end = var_2003_end_0, end_mask = var_2003_end_mask_0, squeeze_mask = var_2003_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2003_cast_fp16")]; + bool var_2004_transpose_x_0 = const()[name = string("op_2004_transpose_x_0"), val = bool(false)]; + bool var_2004_transpose_y_0 = const()[name = string("op_2004_transpose_y_0"), val = bool(false)]; + tensor var_2004_cast_fp16 = matmul(transpose_x = var_2004_transpose_x_0, transpose_y = var_2004_transpose_y_0, x = var_2003_cast_fp16, y = v_new_73_cast_fp16)[name = string("op_2004_cast_fp16")]; + tensor var_2005_cast_fp16 = add(x = var_2001_cast_fp16, y = var_2004_cast_fp16)[name = string("op_2005_cast_fp16")]; + tensor var_2007_begin_0 = const()[name = string("op_2007_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2007_end_0 = const()[name = string("op_2007_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_2007_end_mask_0 = const()[name = string("op_2007_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2007_squeeze_mask_0 = const()[name = string("op_2007_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2007_cast_fp16 = slice_by_index(begin = var_2007_begin_0, end = var_2007_end_0, end_mask = var_2007_end_mask_0, squeeze_mask = var_2007_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2007_cast_fp16")]; + tensor var_2008_cast_fp16 = mul(x = state_73_cast_fp16, y = var_2007_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2010_begin_0 = const()[name = string("op_2010_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2010_end_0 = const()[name = string("op_2010_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2010_end_mask_0 = const()[name = string("op_2010_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2010_squeeze_mask_0 = const()[name = string("op_2010_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2010_cast_fp16 = slice_by_index(begin = var_2010_begin_0, end = var_2010_end_0, end_mask = var_2010_end_mask_0, squeeze_mask = var_2010_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2010_cast_fp16")]; + bool var_2012_transpose_x_1 = const()[name = string("op_2012_transpose_x_1"), val = bool(true)]; + bool var_2012_transpose_y_1 = const()[name = string("op_2012_transpose_y_1"), val = bool(false)]; + tensor var_2012_cast_fp16 = matmul(transpose_x = var_2012_transpose_x_1, transpose_y = var_2012_transpose_y_1, x = var_2010_cast_fp16, y = v_new_73_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor state_75_cast_fp16 = add(x = var_2008_cast_fp16, y = var_2012_cast_fp16)[name = string("state_75_cast_fp16")]; + tensor var_2015_begin_0 = const()[name = string("op_2015_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2015_end_0 = const()[name = string("op_2015_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2015_end_mask_0 = const()[name = string("op_2015_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2015_squeeze_mask_0 = const()[name = string("op_2015_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2015_cast_fp16 = slice_by_index(begin = var_2015_begin_0, end = var_2015_end_0, end_mask = var_2015_end_mask_0, squeeze_mask = var_2015_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2015_cast_fp16")]; + tensor var_2017_begin_0 = const()[name = string("op_2017_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2017_end_0 = const()[name = string("op_2017_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2017_end_mask_0 = const()[name = string("op_2017_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2017_squeeze_mask_0 = const()[name = string("op_2017_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2017_cast_fp16 = slice_by_index(begin = var_2017_begin_0, end = var_2017_end_0, end_mask = var_2017_end_mask_0, squeeze_mask = var_2017_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2017_cast_fp16")]; + bool var_2018_transpose_x_0 = const()[name = string("op_2018_transpose_x_0"), val = bool(false)]; + bool var_2018_transpose_y_0 = const()[name = string("op_2018_transpose_y_0"), val = bool(false)]; + tensor var_2018_cast_fp16 = matmul(transpose_x = var_2018_transpose_x_0, transpose_y = var_2018_transpose_y_0, x = var_2017_cast_fp16, y = state_75_cast_fp16)[name = string("op_2018_cast_fp16")]; + tensor v_new_75_cast_fp16 = sub(x = var_2015_cast_fp16, y = var_2018_cast_fp16)[name = string("v_new_75_cast_fp16")]; + tensor var_2021_begin_0 = const()[name = string("op_2021_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2021_end_0 = const()[name = string("op_2021_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2021_end_mask_0 = const()[name = string("op_2021_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2021_squeeze_mask_0 = const()[name = string("op_2021_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2021_cast_fp16 = slice_by_index(begin = var_2021_begin_0, end = var_2021_end_0, end_mask = var_2021_end_mask_0, squeeze_mask = var_2021_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2021_cast_fp16")]; + bool var_2022_transpose_x_0 = const()[name = string("op_2022_transpose_x_0"), val = bool(false)]; + bool var_2022_transpose_y_0 = const()[name = string("op_2022_transpose_y_0"), val = bool(false)]; + tensor var_2022_cast_fp16 = matmul(transpose_x = var_2022_transpose_x_0, transpose_y = var_2022_transpose_y_0, x = var_2021_cast_fp16, y = state_75_cast_fp16)[name = string("op_2022_cast_fp16")]; + tensor var_2024_begin_0 = const()[name = string("op_2024_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2024_end_0 = const()[name = string("op_2024_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_2024_end_mask_0 = const()[name = string("op_2024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2024_squeeze_mask_0 = const()[name = string("op_2024_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2024_cast_fp16 = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, squeeze_mask = var_2024_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2024_cast_fp16")]; + bool var_2025_transpose_x_0 = const()[name = string("op_2025_transpose_x_0"), val = bool(false)]; + bool var_2025_transpose_y_0 = const()[name = string("op_2025_transpose_y_0"), val = bool(false)]; + tensor var_2025_cast_fp16 = matmul(transpose_x = var_2025_transpose_x_0, transpose_y = var_2025_transpose_y_0, x = var_2024_cast_fp16, y = v_new_75_cast_fp16)[name = string("op_2025_cast_fp16")]; + tensor var_2026_cast_fp16 = add(x = var_2022_cast_fp16, y = var_2025_cast_fp16)[name = string("op_2026_cast_fp16")]; + tensor var_2028_begin_0 = const()[name = string("op_2028_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2028_end_0 = const()[name = string("op_2028_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_2028_end_mask_0 = const()[name = string("op_2028_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2028_squeeze_mask_0 = const()[name = string("op_2028_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, squeeze_mask = var_2028_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2029_cast_fp16 = mul(x = state_75_cast_fp16, y = var_2028_cast_fp16)[name = string("op_2029_cast_fp16")]; + tensor var_2031_begin_0 = const()[name = string("op_2031_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2031_end_0 = const()[name = string("op_2031_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2031_end_mask_0 = const()[name = string("op_2031_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2031_squeeze_mask_0 = const()[name = string("op_2031_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2031_cast_fp16 = slice_by_index(begin = var_2031_begin_0, end = var_2031_end_0, end_mask = var_2031_end_mask_0, squeeze_mask = var_2031_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2031_cast_fp16")]; + bool var_2033_transpose_x_1 = const()[name = string("op_2033_transpose_x_1"), val = bool(true)]; + bool var_2033_transpose_y_1 = const()[name = string("op_2033_transpose_y_1"), val = bool(false)]; + tensor var_2033_cast_fp16 = matmul(transpose_x = var_2033_transpose_x_1, transpose_y = var_2033_transpose_y_1, x = var_2031_cast_fp16, y = v_new_75_cast_fp16)[name = string("op_2033_cast_fp16")]; + tensor state_77_cast_fp16 = add(x = var_2029_cast_fp16, y = var_2033_cast_fp16)[name = string("state_77_cast_fp16")]; + tensor var_2036_begin_0 = const()[name = string("op_2036_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2036_end_0 = const()[name = string("op_2036_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2036_end_mask_0 = const()[name = string("op_2036_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2036_squeeze_mask_0 = const()[name = string("op_2036_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2036_cast_fp16 = slice_by_index(begin = var_2036_begin_0, end = var_2036_end_0, end_mask = var_2036_end_mask_0, squeeze_mask = var_2036_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2036_cast_fp16")]; + tensor var_2038_begin_0 = const()[name = string("op_2038_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2038_end_0 = const()[name = string("op_2038_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2038_end_mask_0 = const()[name = string("op_2038_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2038_squeeze_mask_0 = const()[name = string("op_2038_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2038_cast_fp16 = slice_by_index(begin = var_2038_begin_0, end = var_2038_end_0, end_mask = var_2038_end_mask_0, squeeze_mask = var_2038_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2038_cast_fp16")]; + bool var_2039_transpose_x_0 = const()[name = string("op_2039_transpose_x_0"), val = bool(false)]; + bool var_2039_transpose_y_0 = const()[name = string("op_2039_transpose_y_0"), val = bool(false)]; + tensor var_2039_cast_fp16 = matmul(transpose_x = var_2039_transpose_x_0, transpose_y = var_2039_transpose_y_0, x = var_2038_cast_fp16, y = state_77_cast_fp16)[name = string("op_2039_cast_fp16")]; + tensor v_new_77_cast_fp16 = sub(x = var_2036_cast_fp16, y = var_2039_cast_fp16)[name = string("v_new_77_cast_fp16")]; + tensor var_2042_begin_0 = const()[name = string("op_2042_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2042_end_0 = const()[name = string("op_2042_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2042_end_mask_0 = const()[name = string("op_2042_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2042_squeeze_mask_0 = const()[name = string("op_2042_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2042_cast_fp16 = slice_by_index(begin = var_2042_begin_0, end = var_2042_end_0, end_mask = var_2042_end_mask_0, squeeze_mask = var_2042_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2042_cast_fp16")]; + bool var_2043_transpose_x_0 = const()[name = string("op_2043_transpose_x_0"), val = bool(false)]; + bool var_2043_transpose_y_0 = const()[name = string("op_2043_transpose_y_0"), val = bool(false)]; + tensor var_2043_cast_fp16 = matmul(transpose_x = var_2043_transpose_x_0, transpose_y = var_2043_transpose_y_0, x = var_2042_cast_fp16, y = state_77_cast_fp16)[name = string("op_2043_cast_fp16")]; + tensor var_2045_begin_0 = const()[name = string("op_2045_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2045_end_0 = const()[name = string("op_2045_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_2045_end_mask_0 = const()[name = string("op_2045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2045_squeeze_mask_0 = const()[name = string("op_2045_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2045_cast_fp16 = slice_by_index(begin = var_2045_begin_0, end = var_2045_end_0, end_mask = var_2045_end_mask_0, squeeze_mask = var_2045_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2045_cast_fp16")]; + bool var_2046_transpose_x_0 = const()[name = string("op_2046_transpose_x_0"), val = bool(false)]; + bool var_2046_transpose_y_0 = const()[name = string("op_2046_transpose_y_0"), val = bool(false)]; + tensor var_2046_cast_fp16 = matmul(transpose_x = var_2046_transpose_x_0, transpose_y = var_2046_transpose_y_0, x = var_2045_cast_fp16, y = v_new_77_cast_fp16)[name = string("op_2046_cast_fp16")]; + tensor var_2047_cast_fp16 = add(x = var_2043_cast_fp16, y = var_2046_cast_fp16)[name = string("op_2047_cast_fp16")]; + tensor var_2049_begin_0 = const()[name = string("op_2049_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2049_end_0 = const()[name = string("op_2049_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_2049_end_mask_0 = const()[name = string("op_2049_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2049_squeeze_mask_0 = const()[name = string("op_2049_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2049_cast_fp16 = slice_by_index(begin = var_2049_begin_0, end = var_2049_end_0, end_mask = var_2049_end_mask_0, squeeze_mask = var_2049_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2049_cast_fp16")]; + tensor var_2050_cast_fp16 = mul(x = state_77_cast_fp16, y = var_2049_cast_fp16)[name = string("op_2050_cast_fp16")]; + tensor var_2052_begin_0 = const()[name = string("op_2052_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2052_end_0 = const()[name = string("op_2052_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2052_end_mask_0 = const()[name = string("op_2052_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2052_squeeze_mask_0 = const()[name = string("op_2052_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2052_cast_fp16 = slice_by_index(begin = var_2052_begin_0, end = var_2052_end_0, end_mask = var_2052_end_mask_0, squeeze_mask = var_2052_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2052_cast_fp16")]; + bool var_2054_transpose_x_1 = const()[name = string("op_2054_transpose_x_1"), val = bool(true)]; + bool var_2054_transpose_y_1 = const()[name = string("op_2054_transpose_y_1"), val = bool(false)]; + tensor var_2054_cast_fp16 = matmul(transpose_x = var_2054_transpose_x_1, transpose_y = var_2054_transpose_y_1, x = var_2052_cast_fp16, y = v_new_77_cast_fp16)[name = string("op_2054_cast_fp16")]; + tensor state_79_cast_fp16 = add(x = var_2050_cast_fp16, y = var_2054_cast_fp16)[name = string("state_79_cast_fp16")]; + tensor var_2057_begin_0 = const()[name = string("op_2057_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2057_end_0 = const()[name = string("op_2057_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2057_end_mask_0 = const()[name = string("op_2057_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2057_squeeze_mask_0 = const()[name = string("op_2057_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2057_cast_fp16 = slice_by_index(begin = var_2057_begin_0, end = var_2057_end_0, end_mask = var_2057_end_mask_0, squeeze_mask = var_2057_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2057_cast_fp16")]; + tensor var_2059_begin_0 = const()[name = string("op_2059_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2059_end_0 = const()[name = string("op_2059_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2059_end_mask_0 = const()[name = string("op_2059_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2059_squeeze_mask_0 = const()[name = string("op_2059_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2059_cast_fp16 = slice_by_index(begin = var_2059_begin_0, end = var_2059_end_0, end_mask = var_2059_end_mask_0, squeeze_mask = var_2059_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2059_cast_fp16")]; + bool var_2060_transpose_x_0 = const()[name = string("op_2060_transpose_x_0"), val = bool(false)]; + bool var_2060_transpose_y_0 = const()[name = string("op_2060_transpose_y_0"), val = bool(false)]; + tensor var_2060_cast_fp16 = matmul(transpose_x = var_2060_transpose_x_0, transpose_y = var_2060_transpose_y_0, x = var_2059_cast_fp16, y = state_79_cast_fp16)[name = string("op_2060_cast_fp16")]; + tensor v_new_79_cast_fp16 = sub(x = var_2057_cast_fp16, y = var_2060_cast_fp16)[name = string("v_new_79_cast_fp16")]; + tensor var_2063_begin_0 = const()[name = string("op_2063_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2063_end_0 = const()[name = string("op_2063_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2063_end_mask_0 = const()[name = string("op_2063_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2063_squeeze_mask_0 = const()[name = string("op_2063_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2063_cast_fp16 = slice_by_index(begin = var_2063_begin_0, end = var_2063_end_0, end_mask = var_2063_end_mask_0, squeeze_mask = var_2063_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2063_cast_fp16")]; + bool var_2064_transpose_x_0 = const()[name = string("op_2064_transpose_x_0"), val = bool(false)]; + bool var_2064_transpose_y_0 = const()[name = string("op_2064_transpose_y_0"), val = bool(false)]; + tensor var_2064_cast_fp16 = matmul(transpose_x = var_2064_transpose_x_0, transpose_y = var_2064_transpose_y_0, x = var_2063_cast_fp16, y = state_79_cast_fp16)[name = string("op_2064_cast_fp16")]; + tensor var_2066_begin_0 = const()[name = string("op_2066_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2066_end_0 = const()[name = string("op_2066_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_2066_end_mask_0 = const()[name = string("op_2066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2066_squeeze_mask_0 = const()[name = string("op_2066_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2066_cast_fp16 = slice_by_index(begin = var_2066_begin_0, end = var_2066_end_0, end_mask = var_2066_end_mask_0, squeeze_mask = var_2066_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2066_cast_fp16")]; + bool var_2067_transpose_x_0 = const()[name = string("op_2067_transpose_x_0"), val = bool(false)]; + bool var_2067_transpose_y_0 = const()[name = string("op_2067_transpose_y_0"), val = bool(false)]; + tensor var_2067_cast_fp16 = matmul(transpose_x = var_2067_transpose_x_0, transpose_y = var_2067_transpose_y_0, x = var_2066_cast_fp16, y = v_new_79_cast_fp16)[name = string("op_2067_cast_fp16")]; + tensor var_2068_cast_fp16 = add(x = var_2064_cast_fp16, y = var_2067_cast_fp16)[name = string("op_2068_cast_fp16")]; + tensor var_2070_begin_0 = const()[name = string("op_2070_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2070_end_0 = const()[name = string("op_2070_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_2070_end_mask_0 = const()[name = string("op_2070_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2070_squeeze_mask_0 = const()[name = string("op_2070_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2070_cast_fp16 = slice_by_index(begin = var_2070_begin_0, end = var_2070_end_0, end_mask = var_2070_end_mask_0, squeeze_mask = var_2070_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2070_cast_fp16")]; + tensor var_2071_cast_fp16 = mul(x = state_79_cast_fp16, y = var_2070_cast_fp16)[name = string("op_2071_cast_fp16")]; + tensor var_2073_begin_0 = const()[name = string("op_2073_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2073_end_0 = const()[name = string("op_2073_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2073_end_mask_0 = const()[name = string("op_2073_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2073_squeeze_mask_0 = const()[name = string("op_2073_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2073_cast_fp16 = slice_by_index(begin = var_2073_begin_0, end = var_2073_end_0, end_mask = var_2073_end_mask_0, squeeze_mask = var_2073_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2073_cast_fp16")]; + bool var_2075_transpose_x_1 = const()[name = string("op_2075_transpose_x_1"), val = bool(true)]; + bool var_2075_transpose_y_1 = const()[name = string("op_2075_transpose_y_1"), val = bool(false)]; + tensor var_2075_cast_fp16 = matmul(transpose_x = var_2075_transpose_x_1, transpose_y = var_2075_transpose_y_1, x = var_2073_cast_fp16, y = v_new_79_cast_fp16)[name = string("op_2075_cast_fp16")]; + tensor state_81_cast_fp16 = add(x = var_2071_cast_fp16, y = var_2075_cast_fp16)[name = string("state_81_cast_fp16")]; + tensor var_2078_begin_0 = const()[name = string("op_2078_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2078_end_0 = const()[name = string("op_2078_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_2078_end_mask_0 = const()[name = string("op_2078_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2078_squeeze_mask_0 = const()[name = string("op_2078_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2078_cast_fp16 = slice_by_index(begin = var_2078_begin_0, end = var_2078_end_0, end_mask = var_2078_end_mask_0, squeeze_mask = var_2078_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2078_cast_fp16")]; + tensor var_2080_begin_0 = const()[name = string("op_2080_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2080_end_0 = const()[name = string("op_2080_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_2080_end_mask_0 = const()[name = string("op_2080_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2080_squeeze_mask_0 = const()[name = string("op_2080_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2080_cast_fp16 = slice_by_index(begin = var_2080_begin_0, end = var_2080_end_0, end_mask = var_2080_end_mask_0, squeeze_mask = var_2080_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2080_cast_fp16")]; + bool var_2081_transpose_x_0 = const()[name = string("op_2081_transpose_x_0"), val = bool(false)]; + bool var_2081_transpose_y_0 = const()[name = string("op_2081_transpose_y_0"), val = bool(false)]; + tensor var_2081_cast_fp16 = matmul(transpose_x = var_2081_transpose_x_0, transpose_y = var_2081_transpose_y_0, x = var_2080_cast_fp16, y = state_81_cast_fp16)[name = string("op_2081_cast_fp16")]; + tensor v_new_81_cast_fp16 = sub(x = var_2078_cast_fp16, y = var_2081_cast_fp16)[name = string("v_new_81_cast_fp16")]; + tensor var_2084_begin_0 = const()[name = string("op_2084_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2084_end_0 = const()[name = string("op_2084_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_2084_end_mask_0 = const()[name = string("op_2084_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2084_squeeze_mask_0 = const()[name = string("op_2084_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2084_cast_fp16 = slice_by_index(begin = var_2084_begin_0, end = var_2084_end_0, end_mask = var_2084_end_mask_0, squeeze_mask = var_2084_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2084_cast_fp16")]; + bool var_2085_transpose_x_0 = const()[name = string("op_2085_transpose_x_0"), val = bool(false)]; + bool var_2085_transpose_y_0 = const()[name = string("op_2085_transpose_y_0"), val = bool(false)]; + tensor var_2085_cast_fp16 = matmul(transpose_x = var_2085_transpose_x_0, transpose_y = var_2085_transpose_y_0, x = var_2084_cast_fp16, y = state_81_cast_fp16)[name = string("op_2085_cast_fp16")]; + tensor var_2087_begin_0 = const()[name = string("op_2087_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2087_end_0 = const()[name = string("op_2087_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_2087_end_mask_0 = const()[name = string("op_2087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2087_squeeze_mask_0 = const()[name = string("op_2087_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2087_cast_fp16 = slice_by_index(begin = var_2087_begin_0, end = var_2087_end_0, end_mask = var_2087_end_mask_0, squeeze_mask = var_2087_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2087_cast_fp16")]; + bool var_2088_transpose_x_0 = const()[name = string("op_2088_transpose_x_0"), val = bool(false)]; + bool var_2088_transpose_y_0 = const()[name = string("op_2088_transpose_y_0"), val = bool(false)]; + tensor var_2088_cast_fp16 = matmul(transpose_x = var_2088_transpose_x_0, transpose_y = var_2088_transpose_y_0, x = var_2087_cast_fp16, y = v_new_81_cast_fp16)[name = string("op_2088_cast_fp16")]; + tensor var_2089_cast_fp16 = add(x = var_2085_cast_fp16, y = var_2088_cast_fp16)[name = string("op_2089_cast_fp16")]; + tensor var_2091_begin_0 = const()[name = string("op_2091_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2091_end_0 = const()[name = string("op_2091_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_2091_end_mask_0 = const()[name = string("op_2091_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2091_squeeze_mask_0 = const()[name = string("op_2091_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2091_cast_fp16 = slice_by_index(begin = var_2091_begin_0, end = var_2091_end_0, end_mask = var_2091_end_mask_0, squeeze_mask = var_2091_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2091_cast_fp16")]; + tensor var_2092_cast_fp16 = mul(x = state_81_cast_fp16, y = var_2091_cast_fp16)[name = string("op_2092_cast_fp16")]; + tensor var_2094_begin_0 = const()[name = string("op_2094_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_2094_end_0 = const()[name = string("op_2094_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_2094_end_mask_0 = const()[name = string("op_2094_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2094_squeeze_mask_0 = const()[name = string("op_2094_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2094_cast_fp16 = slice_by_index(begin = var_2094_begin_0, end = var_2094_end_0, end_mask = var_2094_end_mask_0, squeeze_mask = var_2094_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2094_cast_fp16")]; + bool var_2096_transpose_x_1 = const()[name = string("op_2096_transpose_x_1"), val = bool(true)]; + bool var_2096_transpose_y_1 = const()[name = string("op_2096_transpose_y_1"), val = bool(false)]; + tensor var_2096_cast_fp16 = matmul(transpose_x = var_2096_transpose_x_1, transpose_y = var_2096_transpose_y_1, x = var_2094_cast_fp16, y = v_new_81_cast_fp16)[name = string("op_2096_cast_fp16")]; + tensor state_83_cast_fp16 = add(x = var_2092_cast_fp16, y = var_2096_cast_fp16)[name = string("state_83_cast_fp16")]; + tensor var_2099_begin_0 = const()[name = string("op_2099_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2099_end_0 = const()[name = string("op_2099_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_2099_end_mask_0 = const()[name = string("op_2099_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2099_squeeze_mask_0 = const()[name = string("op_2099_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2099_cast_fp16 = slice_by_index(begin = var_2099_begin_0, end = var_2099_end_0, end_mask = var_2099_end_mask_0, squeeze_mask = var_2099_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2099_cast_fp16")]; + tensor var_2101_begin_0 = const()[name = string("op_2101_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2101_end_0 = const()[name = string("op_2101_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_2101_end_mask_0 = const()[name = string("op_2101_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2101_squeeze_mask_0 = const()[name = string("op_2101_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2101_cast_fp16 = slice_by_index(begin = var_2101_begin_0, end = var_2101_end_0, end_mask = var_2101_end_mask_0, squeeze_mask = var_2101_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2101_cast_fp16")]; + bool var_2102_transpose_x_0 = const()[name = string("op_2102_transpose_x_0"), val = bool(false)]; + bool var_2102_transpose_y_0 = const()[name = string("op_2102_transpose_y_0"), val = bool(false)]; + tensor var_2102_cast_fp16 = matmul(transpose_x = var_2102_transpose_x_0, transpose_y = var_2102_transpose_y_0, x = var_2101_cast_fp16, y = state_83_cast_fp16)[name = string("op_2102_cast_fp16")]; + tensor v_new_83_cast_fp16 = sub(x = var_2099_cast_fp16, y = var_2102_cast_fp16)[name = string("v_new_83_cast_fp16")]; + tensor var_2105_begin_0 = const()[name = string("op_2105_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2105_end_0 = const()[name = string("op_2105_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_2105_end_mask_0 = const()[name = string("op_2105_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2105_squeeze_mask_0 = const()[name = string("op_2105_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2105_cast_fp16 = slice_by_index(begin = var_2105_begin_0, end = var_2105_end_0, end_mask = var_2105_end_mask_0, squeeze_mask = var_2105_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2105_cast_fp16")]; + bool var_2106_transpose_x_0 = const()[name = string("op_2106_transpose_x_0"), val = bool(false)]; + bool var_2106_transpose_y_0 = const()[name = string("op_2106_transpose_y_0"), val = bool(false)]; + tensor var_2106_cast_fp16 = matmul(transpose_x = var_2106_transpose_x_0, transpose_y = var_2106_transpose_y_0, x = var_2105_cast_fp16, y = state_83_cast_fp16)[name = string("op_2106_cast_fp16")]; + tensor var_2108_begin_0 = const()[name = string("op_2108_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2108_end_0 = const()[name = string("op_2108_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_2108_end_mask_0 = const()[name = string("op_2108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2108_squeeze_mask_0 = const()[name = string("op_2108_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2108_cast_fp16 = slice_by_index(begin = var_2108_begin_0, end = var_2108_end_0, end_mask = var_2108_end_mask_0, squeeze_mask = var_2108_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2108_cast_fp16")]; + bool var_2109_transpose_x_0 = const()[name = string("op_2109_transpose_x_0"), val = bool(false)]; + bool var_2109_transpose_y_0 = const()[name = string("op_2109_transpose_y_0"), val = bool(false)]; + tensor var_2109_cast_fp16 = matmul(transpose_x = var_2109_transpose_x_0, transpose_y = var_2109_transpose_y_0, x = var_2108_cast_fp16, y = v_new_83_cast_fp16)[name = string("op_2109_cast_fp16")]; + tensor var_2110_cast_fp16 = add(x = var_2106_cast_fp16, y = var_2109_cast_fp16)[name = string("op_2110_cast_fp16")]; + tensor var_2112_begin_0 = const()[name = string("op_2112_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2112_end_0 = const()[name = string("op_2112_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_2112_end_mask_0 = const()[name = string("op_2112_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2112_squeeze_mask_0 = const()[name = string("op_2112_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2112_cast_fp16 = slice_by_index(begin = var_2112_begin_0, end = var_2112_end_0, end_mask = var_2112_end_mask_0, squeeze_mask = var_2112_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2112_cast_fp16")]; + tensor var_2113_cast_fp16 = mul(x = state_83_cast_fp16, y = var_2112_cast_fp16)[name = string("op_2113_cast_fp16")]; + tensor var_2115_begin_0 = const()[name = string("op_2115_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_2115_end_0 = const()[name = string("op_2115_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_2115_end_mask_0 = const()[name = string("op_2115_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2115_squeeze_mask_0 = const()[name = string("op_2115_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2115_cast_fp16 = slice_by_index(begin = var_2115_begin_0, end = var_2115_end_0, end_mask = var_2115_end_mask_0, squeeze_mask = var_2115_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2115_cast_fp16")]; + bool var_2117_transpose_x_1 = const()[name = string("op_2117_transpose_x_1"), val = bool(true)]; + bool var_2117_transpose_y_1 = const()[name = string("op_2117_transpose_y_1"), val = bool(false)]; + tensor var_2117_cast_fp16 = matmul(transpose_x = var_2117_transpose_x_1, transpose_y = var_2117_transpose_y_1, x = var_2115_cast_fp16, y = v_new_83_cast_fp16)[name = string("op_2117_cast_fp16")]; + tensor state_85_cast_fp16 = add(x = var_2113_cast_fp16, y = var_2117_cast_fp16)[name = string("state_85_cast_fp16")]; + tensor var_2120_begin_0 = const()[name = string("op_2120_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2120_end_0 = const()[name = string("op_2120_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_2120_end_mask_0 = const()[name = string("op_2120_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2120_squeeze_mask_0 = const()[name = string("op_2120_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2120_cast_fp16 = slice_by_index(begin = var_2120_begin_0, end = var_2120_end_0, end_mask = var_2120_end_mask_0, squeeze_mask = var_2120_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2120_cast_fp16")]; + tensor var_2122_begin_0 = const()[name = string("op_2122_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2122_end_0 = const()[name = string("op_2122_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_2122_end_mask_0 = const()[name = string("op_2122_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2122_squeeze_mask_0 = const()[name = string("op_2122_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2122_cast_fp16 = slice_by_index(begin = var_2122_begin_0, end = var_2122_end_0, end_mask = var_2122_end_mask_0, squeeze_mask = var_2122_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2122_cast_fp16")]; + bool var_2123_transpose_x_0 = const()[name = string("op_2123_transpose_x_0"), val = bool(false)]; + bool var_2123_transpose_y_0 = const()[name = string("op_2123_transpose_y_0"), val = bool(false)]; + tensor var_2123_cast_fp16 = matmul(transpose_x = var_2123_transpose_x_0, transpose_y = var_2123_transpose_y_0, x = var_2122_cast_fp16, y = state_85_cast_fp16)[name = string("op_2123_cast_fp16")]; + tensor v_new_85_cast_fp16 = sub(x = var_2120_cast_fp16, y = var_2123_cast_fp16)[name = string("v_new_85_cast_fp16")]; + tensor var_2126_begin_0 = const()[name = string("op_2126_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2126_end_0 = const()[name = string("op_2126_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_2126_end_mask_0 = const()[name = string("op_2126_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2126_squeeze_mask_0 = const()[name = string("op_2126_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2126_cast_fp16 = slice_by_index(begin = var_2126_begin_0, end = var_2126_end_0, end_mask = var_2126_end_mask_0, squeeze_mask = var_2126_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2126_cast_fp16")]; + bool var_2127_transpose_x_0 = const()[name = string("op_2127_transpose_x_0"), val = bool(false)]; + bool var_2127_transpose_y_0 = const()[name = string("op_2127_transpose_y_0"), val = bool(false)]; + tensor var_2127_cast_fp16 = matmul(transpose_x = var_2127_transpose_x_0, transpose_y = var_2127_transpose_y_0, x = var_2126_cast_fp16, y = state_85_cast_fp16)[name = string("op_2127_cast_fp16")]; + tensor var_2129_begin_0 = const()[name = string("op_2129_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2129_end_0 = const()[name = string("op_2129_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_2129_end_mask_0 = const()[name = string("op_2129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2129_squeeze_mask_0 = const()[name = string("op_2129_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2129_cast_fp16 = slice_by_index(begin = var_2129_begin_0, end = var_2129_end_0, end_mask = var_2129_end_mask_0, squeeze_mask = var_2129_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2129_cast_fp16")]; + bool var_2130_transpose_x_0 = const()[name = string("op_2130_transpose_x_0"), val = bool(false)]; + bool var_2130_transpose_y_0 = const()[name = string("op_2130_transpose_y_0"), val = bool(false)]; + tensor var_2130_cast_fp16 = matmul(transpose_x = var_2130_transpose_x_0, transpose_y = var_2130_transpose_y_0, x = var_2129_cast_fp16, y = v_new_85_cast_fp16)[name = string("op_2130_cast_fp16")]; + tensor var_2131_cast_fp16 = add(x = var_2127_cast_fp16, y = var_2130_cast_fp16)[name = string("op_2131_cast_fp16")]; + tensor var_2133_begin_0 = const()[name = string("op_2133_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2133_end_0 = const()[name = string("op_2133_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_2133_end_mask_0 = const()[name = string("op_2133_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2133_squeeze_mask_0 = const()[name = string("op_2133_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2133_cast_fp16 = slice_by_index(begin = var_2133_begin_0, end = var_2133_end_0, end_mask = var_2133_end_mask_0, squeeze_mask = var_2133_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2133_cast_fp16")]; + tensor var_2134_cast_fp16 = mul(x = state_85_cast_fp16, y = var_2133_cast_fp16)[name = string("op_2134_cast_fp16")]; + tensor var_2136_begin_0 = const()[name = string("op_2136_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_2136_end_0 = const()[name = string("op_2136_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_2136_end_mask_0 = const()[name = string("op_2136_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2136_squeeze_mask_0 = const()[name = string("op_2136_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2136_cast_fp16 = slice_by_index(begin = var_2136_begin_0, end = var_2136_end_0, end_mask = var_2136_end_mask_0, squeeze_mask = var_2136_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2136_cast_fp16")]; + bool var_2138_transpose_x_1 = const()[name = string("op_2138_transpose_x_1"), val = bool(true)]; + bool var_2138_transpose_y_1 = const()[name = string("op_2138_transpose_y_1"), val = bool(false)]; + tensor var_2138_cast_fp16 = matmul(transpose_x = var_2138_transpose_x_1, transpose_y = var_2138_transpose_y_1, x = var_2136_cast_fp16, y = v_new_85_cast_fp16)[name = string("op_2138_cast_fp16")]; + tensor state_87_cast_fp16 = add(x = var_2134_cast_fp16, y = var_2138_cast_fp16)[name = string("state_87_cast_fp16")]; + tensor var_2141_begin_0 = const()[name = string("op_2141_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2141_end_0 = const()[name = string("op_2141_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_2141_end_mask_0 = const()[name = string("op_2141_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2141_squeeze_mask_0 = const()[name = string("op_2141_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2141_cast_fp16 = slice_by_index(begin = var_2141_begin_0, end = var_2141_end_0, end_mask = var_2141_end_mask_0, squeeze_mask = var_2141_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2141_cast_fp16")]; + tensor var_2143_begin_0 = const()[name = string("op_2143_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2143_end_0 = const()[name = string("op_2143_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_2143_end_mask_0 = const()[name = string("op_2143_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2143_squeeze_mask_0 = const()[name = string("op_2143_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2143_cast_fp16 = slice_by_index(begin = var_2143_begin_0, end = var_2143_end_0, end_mask = var_2143_end_mask_0, squeeze_mask = var_2143_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2143_cast_fp16")]; + bool var_2144_transpose_x_0 = const()[name = string("op_2144_transpose_x_0"), val = bool(false)]; + bool var_2144_transpose_y_0 = const()[name = string("op_2144_transpose_y_0"), val = bool(false)]; + tensor var_2144_cast_fp16 = matmul(transpose_x = var_2144_transpose_x_0, transpose_y = var_2144_transpose_y_0, x = var_2143_cast_fp16, y = state_87_cast_fp16)[name = string("op_2144_cast_fp16")]; + tensor v_new_87_cast_fp16 = sub(x = var_2141_cast_fp16, y = var_2144_cast_fp16)[name = string("v_new_87_cast_fp16")]; + tensor var_2147_begin_0 = const()[name = string("op_2147_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2147_end_0 = const()[name = string("op_2147_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_2147_end_mask_0 = const()[name = string("op_2147_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2147_squeeze_mask_0 = const()[name = string("op_2147_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2147_cast_fp16 = slice_by_index(begin = var_2147_begin_0, end = var_2147_end_0, end_mask = var_2147_end_mask_0, squeeze_mask = var_2147_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2147_cast_fp16")]; + bool var_2148_transpose_x_0 = const()[name = string("op_2148_transpose_x_0"), val = bool(false)]; + bool var_2148_transpose_y_0 = const()[name = string("op_2148_transpose_y_0"), val = bool(false)]; + tensor var_2148_cast_fp16 = matmul(transpose_x = var_2148_transpose_x_0, transpose_y = var_2148_transpose_y_0, x = var_2147_cast_fp16, y = state_87_cast_fp16)[name = string("op_2148_cast_fp16")]; + tensor var_2150_begin_0 = const()[name = string("op_2150_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2150_end_0 = const()[name = string("op_2150_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_2150_end_mask_0 = const()[name = string("op_2150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2150_squeeze_mask_0 = const()[name = string("op_2150_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2150_cast_fp16 = slice_by_index(begin = var_2150_begin_0, end = var_2150_end_0, end_mask = var_2150_end_mask_0, squeeze_mask = var_2150_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2150_cast_fp16")]; + bool var_2151_transpose_x_0 = const()[name = string("op_2151_transpose_x_0"), val = bool(false)]; + bool var_2151_transpose_y_0 = const()[name = string("op_2151_transpose_y_0"), val = bool(false)]; + tensor var_2151_cast_fp16 = matmul(transpose_x = var_2151_transpose_x_0, transpose_y = var_2151_transpose_y_0, x = var_2150_cast_fp16, y = v_new_87_cast_fp16)[name = string("op_2151_cast_fp16")]; + tensor var_2152_cast_fp16 = add(x = var_2148_cast_fp16, y = var_2151_cast_fp16)[name = string("op_2152_cast_fp16")]; + tensor var_2154_begin_0 = const()[name = string("op_2154_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2154_end_0 = const()[name = string("op_2154_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_2154_end_mask_0 = const()[name = string("op_2154_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2154_squeeze_mask_0 = const()[name = string("op_2154_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2154_cast_fp16 = slice_by_index(begin = var_2154_begin_0, end = var_2154_end_0, end_mask = var_2154_end_mask_0, squeeze_mask = var_2154_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2154_cast_fp16")]; + tensor var_2155_cast_fp16 = mul(x = state_87_cast_fp16, y = var_2154_cast_fp16)[name = string("op_2155_cast_fp16")]; + tensor var_2157_begin_0 = const()[name = string("op_2157_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_2157_end_0 = const()[name = string("op_2157_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_2157_end_mask_0 = const()[name = string("op_2157_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2157_squeeze_mask_0 = const()[name = string("op_2157_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2157_cast_fp16 = slice_by_index(begin = var_2157_begin_0, end = var_2157_end_0, end_mask = var_2157_end_mask_0, squeeze_mask = var_2157_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2157_cast_fp16")]; + bool var_2159_transpose_x_1 = const()[name = string("op_2159_transpose_x_1"), val = bool(true)]; + bool var_2159_transpose_y_1 = const()[name = string("op_2159_transpose_y_1"), val = bool(false)]; + tensor var_2159_cast_fp16 = matmul(transpose_x = var_2159_transpose_x_1, transpose_y = var_2159_transpose_y_1, x = var_2157_cast_fp16, y = v_new_87_cast_fp16)[name = string("op_2159_cast_fp16")]; + tensor state_89_cast_fp16 = add(x = var_2155_cast_fp16, y = var_2159_cast_fp16)[name = string("state_89_cast_fp16")]; + tensor var_2162_begin_0 = const()[name = string("op_2162_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2162_end_0 = const()[name = string("op_2162_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_2162_end_mask_0 = const()[name = string("op_2162_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2162_squeeze_mask_0 = const()[name = string("op_2162_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2162_cast_fp16 = slice_by_index(begin = var_2162_begin_0, end = var_2162_end_0, end_mask = var_2162_end_mask_0, squeeze_mask = var_2162_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2162_cast_fp16")]; + tensor var_2164_begin_0 = const()[name = string("op_2164_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2164_end_0 = const()[name = string("op_2164_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_2164_end_mask_0 = const()[name = string("op_2164_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2164_squeeze_mask_0 = const()[name = string("op_2164_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2164_cast_fp16 = slice_by_index(begin = var_2164_begin_0, end = var_2164_end_0, end_mask = var_2164_end_mask_0, squeeze_mask = var_2164_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2164_cast_fp16")]; + bool var_2165_transpose_x_0 = const()[name = string("op_2165_transpose_x_0"), val = bool(false)]; + bool var_2165_transpose_y_0 = const()[name = string("op_2165_transpose_y_0"), val = bool(false)]; + tensor var_2165_cast_fp16 = matmul(transpose_x = var_2165_transpose_x_0, transpose_y = var_2165_transpose_y_0, x = var_2164_cast_fp16, y = state_89_cast_fp16)[name = string("op_2165_cast_fp16")]; + tensor v_new_89_cast_fp16 = sub(x = var_2162_cast_fp16, y = var_2165_cast_fp16)[name = string("v_new_89_cast_fp16")]; + tensor var_2168_begin_0 = const()[name = string("op_2168_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2168_end_0 = const()[name = string("op_2168_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_2168_end_mask_0 = const()[name = string("op_2168_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2168_squeeze_mask_0 = const()[name = string("op_2168_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2168_cast_fp16 = slice_by_index(begin = var_2168_begin_0, end = var_2168_end_0, end_mask = var_2168_end_mask_0, squeeze_mask = var_2168_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2168_cast_fp16")]; + bool var_2169_transpose_x_0 = const()[name = string("op_2169_transpose_x_0"), val = bool(false)]; + bool var_2169_transpose_y_0 = const()[name = string("op_2169_transpose_y_0"), val = bool(false)]; + tensor var_2169_cast_fp16 = matmul(transpose_x = var_2169_transpose_x_0, transpose_y = var_2169_transpose_y_0, x = var_2168_cast_fp16, y = state_89_cast_fp16)[name = string("op_2169_cast_fp16")]; + tensor var_2171_begin_0 = const()[name = string("op_2171_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2171_end_0 = const()[name = string("op_2171_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_2171_end_mask_0 = const()[name = string("op_2171_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2171_squeeze_mask_0 = const()[name = string("op_2171_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2171_cast_fp16 = slice_by_index(begin = var_2171_begin_0, end = var_2171_end_0, end_mask = var_2171_end_mask_0, squeeze_mask = var_2171_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2171_cast_fp16")]; + bool var_2172_transpose_x_0 = const()[name = string("op_2172_transpose_x_0"), val = bool(false)]; + bool var_2172_transpose_y_0 = const()[name = string("op_2172_transpose_y_0"), val = bool(false)]; + tensor var_2172_cast_fp16 = matmul(transpose_x = var_2172_transpose_x_0, transpose_y = var_2172_transpose_y_0, x = var_2171_cast_fp16, y = v_new_89_cast_fp16)[name = string("op_2172_cast_fp16")]; + tensor var_2173_cast_fp16 = add(x = var_2169_cast_fp16, y = var_2172_cast_fp16)[name = string("op_2173_cast_fp16")]; + tensor var_2175_begin_0 = const()[name = string("op_2175_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2175_end_0 = const()[name = string("op_2175_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_2175_end_mask_0 = const()[name = string("op_2175_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2175_squeeze_mask_0 = const()[name = string("op_2175_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2175_cast_fp16 = slice_by_index(begin = var_2175_begin_0, end = var_2175_end_0, end_mask = var_2175_end_mask_0, squeeze_mask = var_2175_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2175_cast_fp16")]; + tensor var_2176_cast_fp16 = mul(x = state_89_cast_fp16, y = var_2175_cast_fp16)[name = string("op_2176_cast_fp16")]; + tensor var_2178_begin_0 = const()[name = string("op_2178_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_2178_end_0 = const()[name = string("op_2178_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_2178_end_mask_0 = const()[name = string("op_2178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2178_squeeze_mask_0 = const()[name = string("op_2178_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2178_cast_fp16 = slice_by_index(begin = var_2178_begin_0, end = var_2178_end_0, end_mask = var_2178_end_mask_0, squeeze_mask = var_2178_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2178_cast_fp16")]; + bool var_2180_transpose_x_1 = const()[name = string("op_2180_transpose_x_1"), val = bool(true)]; + bool var_2180_transpose_y_1 = const()[name = string("op_2180_transpose_y_1"), val = bool(false)]; + tensor var_2180_cast_fp16 = matmul(transpose_x = var_2180_transpose_x_1, transpose_y = var_2180_transpose_y_1, x = var_2178_cast_fp16, y = v_new_89_cast_fp16)[name = string("op_2180_cast_fp16")]; + tensor state_91_cast_fp16 = add(x = var_2176_cast_fp16, y = var_2180_cast_fp16)[name = string("state_91_cast_fp16")]; + tensor var_2183_begin_0 = const()[name = string("op_2183_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2183_end_0 = const()[name = string("op_2183_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_2183_end_mask_0 = const()[name = string("op_2183_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2183_squeeze_mask_0 = const()[name = string("op_2183_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2183_cast_fp16 = slice_by_index(begin = var_2183_begin_0, end = var_2183_end_0, end_mask = var_2183_end_mask_0, squeeze_mask = var_2183_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2183_cast_fp16")]; + tensor var_2185_begin_0 = const()[name = string("op_2185_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2185_end_0 = const()[name = string("op_2185_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_2185_end_mask_0 = const()[name = string("op_2185_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2185_squeeze_mask_0 = const()[name = string("op_2185_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2185_cast_fp16 = slice_by_index(begin = var_2185_begin_0, end = var_2185_end_0, end_mask = var_2185_end_mask_0, squeeze_mask = var_2185_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2185_cast_fp16")]; + bool var_2186_transpose_x_0 = const()[name = string("op_2186_transpose_x_0"), val = bool(false)]; + bool var_2186_transpose_y_0 = const()[name = string("op_2186_transpose_y_0"), val = bool(false)]; + tensor var_2186_cast_fp16 = matmul(transpose_x = var_2186_transpose_x_0, transpose_y = var_2186_transpose_y_0, x = var_2185_cast_fp16, y = state_91_cast_fp16)[name = string("op_2186_cast_fp16")]; + tensor v_new_91_cast_fp16 = sub(x = var_2183_cast_fp16, y = var_2186_cast_fp16)[name = string("v_new_91_cast_fp16")]; + tensor var_2189_begin_0 = const()[name = string("op_2189_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2189_end_0 = const()[name = string("op_2189_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_2189_end_mask_0 = const()[name = string("op_2189_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2189_squeeze_mask_0 = const()[name = string("op_2189_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2189_cast_fp16 = slice_by_index(begin = var_2189_begin_0, end = var_2189_end_0, end_mask = var_2189_end_mask_0, squeeze_mask = var_2189_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2189_cast_fp16")]; + bool var_2190_transpose_x_0 = const()[name = string("op_2190_transpose_x_0"), val = bool(false)]; + bool var_2190_transpose_y_0 = const()[name = string("op_2190_transpose_y_0"), val = bool(false)]; + tensor var_2190_cast_fp16 = matmul(transpose_x = var_2190_transpose_x_0, transpose_y = var_2190_transpose_y_0, x = var_2189_cast_fp16, y = state_91_cast_fp16)[name = string("op_2190_cast_fp16")]; + tensor var_2192_begin_0 = const()[name = string("op_2192_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2192_end_0 = const()[name = string("op_2192_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_2192_end_mask_0 = const()[name = string("op_2192_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2192_squeeze_mask_0 = const()[name = string("op_2192_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2192_cast_fp16 = slice_by_index(begin = var_2192_begin_0, end = var_2192_end_0, end_mask = var_2192_end_mask_0, squeeze_mask = var_2192_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2192_cast_fp16")]; + bool var_2193_transpose_x_0 = const()[name = string("op_2193_transpose_x_0"), val = bool(false)]; + bool var_2193_transpose_y_0 = const()[name = string("op_2193_transpose_y_0"), val = bool(false)]; + tensor var_2193_cast_fp16 = matmul(transpose_x = var_2193_transpose_x_0, transpose_y = var_2193_transpose_y_0, x = var_2192_cast_fp16, y = v_new_91_cast_fp16)[name = string("op_2193_cast_fp16")]; + tensor var_2194_cast_fp16 = add(x = var_2190_cast_fp16, y = var_2193_cast_fp16)[name = string("op_2194_cast_fp16")]; + tensor var_2196_begin_0 = const()[name = string("op_2196_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2196_end_0 = const()[name = string("op_2196_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_2196_end_mask_0 = const()[name = string("op_2196_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2196_squeeze_mask_0 = const()[name = string("op_2196_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2196_cast_fp16 = slice_by_index(begin = var_2196_begin_0, end = var_2196_end_0, end_mask = var_2196_end_mask_0, squeeze_mask = var_2196_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2196_cast_fp16")]; + tensor var_2197_cast_fp16 = mul(x = state_91_cast_fp16, y = var_2196_cast_fp16)[name = string("op_2197_cast_fp16")]; + tensor var_2199_begin_0 = const()[name = string("op_2199_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_2199_end_0 = const()[name = string("op_2199_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_2199_end_mask_0 = const()[name = string("op_2199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2199_squeeze_mask_0 = const()[name = string("op_2199_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2199_cast_fp16 = slice_by_index(begin = var_2199_begin_0, end = var_2199_end_0, end_mask = var_2199_end_mask_0, squeeze_mask = var_2199_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2199_cast_fp16")]; + bool var_2201_transpose_x_1 = const()[name = string("op_2201_transpose_x_1"), val = bool(true)]; + bool var_2201_transpose_y_1 = const()[name = string("op_2201_transpose_y_1"), val = bool(false)]; + tensor var_2201_cast_fp16 = matmul(transpose_x = var_2201_transpose_x_1, transpose_y = var_2201_transpose_y_1, x = var_2199_cast_fp16, y = v_new_91_cast_fp16)[name = string("op_2201_cast_fp16")]; + tensor state_93_cast_fp16 = add(x = var_2197_cast_fp16, y = var_2201_cast_fp16)[name = string("state_93_cast_fp16")]; + tensor var_2204_begin_0 = const()[name = string("op_2204_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2204_end_0 = const()[name = string("op_2204_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_2204_end_mask_0 = const()[name = string("op_2204_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2204_squeeze_mask_0 = const()[name = string("op_2204_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2204_cast_fp16 = slice_by_index(begin = var_2204_begin_0, end = var_2204_end_0, end_mask = var_2204_end_mask_0, squeeze_mask = var_2204_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2204_cast_fp16")]; + tensor var_2206_begin_0 = const()[name = string("op_2206_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2206_end_0 = const()[name = string("op_2206_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_2206_end_mask_0 = const()[name = string("op_2206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2206_squeeze_mask_0 = const()[name = string("op_2206_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2206_cast_fp16 = slice_by_index(begin = var_2206_begin_0, end = var_2206_end_0, end_mask = var_2206_end_mask_0, squeeze_mask = var_2206_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2206_cast_fp16")]; + bool var_2207_transpose_x_0 = const()[name = string("op_2207_transpose_x_0"), val = bool(false)]; + bool var_2207_transpose_y_0 = const()[name = string("op_2207_transpose_y_0"), val = bool(false)]; + tensor var_2207_cast_fp16 = matmul(transpose_x = var_2207_transpose_x_0, transpose_y = var_2207_transpose_y_0, x = var_2206_cast_fp16, y = state_93_cast_fp16)[name = string("op_2207_cast_fp16")]; + tensor v_new_93_cast_fp16 = sub(x = var_2204_cast_fp16, y = var_2207_cast_fp16)[name = string("v_new_93_cast_fp16")]; + tensor var_2210_begin_0 = const()[name = string("op_2210_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2210_end_0 = const()[name = string("op_2210_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_2210_end_mask_0 = const()[name = string("op_2210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2210_squeeze_mask_0 = const()[name = string("op_2210_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2210_cast_fp16 = slice_by_index(begin = var_2210_begin_0, end = var_2210_end_0, end_mask = var_2210_end_mask_0, squeeze_mask = var_2210_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2210_cast_fp16")]; + bool var_2211_transpose_x_0 = const()[name = string("op_2211_transpose_x_0"), val = bool(false)]; + bool var_2211_transpose_y_0 = const()[name = string("op_2211_transpose_y_0"), val = bool(false)]; + tensor var_2211_cast_fp16 = matmul(transpose_x = var_2211_transpose_x_0, transpose_y = var_2211_transpose_y_0, x = var_2210_cast_fp16, y = state_93_cast_fp16)[name = string("op_2211_cast_fp16")]; + tensor var_2213_begin_0 = const()[name = string("op_2213_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2213_end_0 = const()[name = string("op_2213_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_2213_end_mask_0 = const()[name = string("op_2213_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2213_squeeze_mask_0 = const()[name = string("op_2213_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2213_cast_fp16 = slice_by_index(begin = var_2213_begin_0, end = var_2213_end_0, end_mask = var_2213_end_mask_0, squeeze_mask = var_2213_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2213_cast_fp16")]; + bool var_2214_transpose_x_0 = const()[name = string("op_2214_transpose_x_0"), val = bool(false)]; + bool var_2214_transpose_y_0 = const()[name = string("op_2214_transpose_y_0"), val = bool(false)]; + tensor var_2214_cast_fp16 = matmul(transpose_x = var_2214_transpose_x_0, transpose_y = var_2214_transpose_y_0, x = var_2213_cast_fp16, y = v_new_93_cast_fp16)[name = string("op_2214_cast_fp16")]; + tensor var_2215_cast_fp16 = add(x = var_2211_cast_fp16, y = var_2214_cast_fp16)[name = string("op_2215_cast_fp16")]; + tensor var_2217_begin_0 = const()[name = string("op_2217_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2217_end_0 = const()[name = string("op_2217_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_2217_end_mask_0 = const()[name = string("op_2217_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2217_squeeze_mask_0 = const()[name = string("op_2217_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2217_cast_fp16 = slice_by_index(begin = var_2217_begin_0, end = var_2217_end_0, end_mask = var_2217_end_mask_0, squeeze_mask = var_2217_squeeze_mask_0, x = chunk_decay_5_cast_fp16)[name = string("op_2217_cast_fp16")]; + tensor var_2218_cast_fp16 = mul(x = state_93_cast_fp16, y = var_2217_cast_fp16)[name = string("op_2218_cast_fp16")]; + tensor var_2220_begin_0 = const()[name = string("op_2220_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_2220_end_0 = const()[name = string("op_2220_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_2220_end_mask_0 = const()[name = string("op_2220_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2220_squeeze_mask_0 = const()[name = string("op_2220_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2220_cast_fp16 = slice_by_index(begin = var_2220_begin_0, end = var_2220_end_0, end_mask = var_2220_end_mask_0, squeeze_mask = var_2220_squeeze_mask_0, x = k_dec_5_cast_fp16)[name = string("op_2220_cast_fp16")]; + bool var_2222_transpose_x_1 = const()[name = string("op_2222_transpose_x_1"), val = bool(true)]; + bool var_2222_transpose_y_1 = const()[name = string("op_2222_transpose_y_1"), val = bool(false)]; + tensor var_2222_cast_fp16 = matmul(transpose_x = var_2222_transpose_x_1, transpose_y = var_2222_transpose_y_1, x = var_2220_cast_fp16, y = v_new_93_cast_fp16)[name = string("op_2222_cast_fp16")]; + tensor state_95_cast_fp16 = add(x = var_2218_cast_fp16, y = var_2222_cast_fp16)[name = string("state_95_cast_fp16")]; + tensor var_2225_begin_0 = const()[name = string("op_2225_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_2225_end_0 = const()[name = string("op_2225_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_2225_end_mask_0 = const()[name = string("op_2225_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2225_squeeze_mask_0 = const()[name = string("op_2225_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2225_cast_fp16 = slice_by_index(begin = var_2225_begin_0, end = var_2225_end_0, end_mask = var_2225_end_mask_0, squeeze_mask = var_2225_squeeze_mask_0, x = new_v_5_cast_fp16)[name = string("op_2225_cast_fp16")]; + tensor var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2227_squeeze_mask_0 = const()[name = string("op_2227_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, squeeze_mask = var_2227_squeeze_mask_0, x = k_cumdecay_5_cast_fp16)[name = string("op_2227_cast_fp16")]; + bool var_2228_transpose_x_0 = const()[name = string("op_2228_transpose_x_0"), val = bool(false)]; + bool var_2228_transpose_y_0 = const()[name = string("op_2228_transpose_y_0"), val = bool(false)]; + tensor var_2228_cast_fp16 = matmul(transpose_x = var_2228_transpose_x_0, transpose_y = var_2228_transpose_y_0, x = var_2227_cast_fp16, y = state_95_cast_fp16)[name = string("op_2228_cast_fp16")]; + tensor v_new_95_cast_fp16 = sub(x = var_2225_cast_fp16, y = var_2228_cast_fp16)[name = string("v_new_95_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2231_squeeze_mask_0 = const()[name = string("op_2231_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, squeeze_mask = var_2231_squeeze_mask_0, x = q_dec_5_cast_fp16)[name = string("op_2231_cast_fp16")]; + bool var_2232_transpose_x_0 = const()[name = string("op_2232_transpose_x_0"), val = bool(false)]; + bool var_2232_transpose_y_0 = const()[name = string("op_2232_transpose_y_0"), val = bool(false)]; + tensor var_2232_cast_fp16 = matmul(transpose_x = var_2232_transpose_x_0, transpose_y = var_2232_transpose_y_0, x = var_2231_cast_fp16, y = state_95_cast_fp16)[name = string("op_2232_cast_fp16")]; + tensor var_2234_begin_0 = const()[name = string("op_2234_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_2234_end_0 = const()[name = string("op_2234_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_2234_end_mask_0 = const()[name = string("op_2234_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2234_squeeze_mask_0 = const()[name = string("op_2234_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2234_cast_fp16 = slice_by_index(begin = var_2234_begin_0, end = var_2234_end_0, end_mask = var_2234_end_mask_0, squeeze_mask = var_2234_squeeze_mask_0, x = attn_intra_5_cast_fp16)[name = string("op_2234_cast_fp16")]; + bool var_2235_transpose_x_0 = const()[name = string("op_2235_transpose_x_0"), val = bool(false)]; + bool var_2235_transpose_y_0 = const()[name = string("op_2235_transpose_y_0"), val = bool(false)]; + tensor var_2235_cast_fp16 = matmul(transpose_x = var_2235_transpose_x_0, transpose_y = var_2235_transpose_y_0, x = var_2234_cast_fp16, y = v_new_95_cast_fp16)[name = string("op_2235_cast_fp16")]; + tensor var_2236_cast_fp16 = add(x = var_2232_cast_fp16, y = var_2235_cast_fp16)[name = string("op_2236_cast_fp16")]; + int32 var_2238_axis_0 = const()[name = string("op_2238_axis_0"), val = int32(1)]; + tensor var_2238_cast_fp16 = stack(axis = var_2238_axis_0, values = (var_1921_cast_fp16, var_1942_cast_fp16, var_1963_cast_fp16, var_1984_cast_fp16, var_2005_cast_fp16, var_2026_cast_fp16, var_2047_cast_fp16, var_2068_cast_fp16, var_2089_cast_fp16, var_2110_cast_fp16, var_2131_cast_fp16, var_2152_cast_fp16, var_2173_cast_fp16, var_2194_cast_fp16, var_2215_cast_fp16, var_2236_cast_fp16))[name = string("op_2238_cast_fp16")]; + tensor var_2239 = const()[name = string("op_2239"), val = tensor([32, 1024, 128])]; + tensor var_2240_cast_fp16 = reshape(shape = var_2239, x = var_2238_cast_fp16)[name = string("op_2240_cast_fp16")]; + tensor var_2241 = const()[name = string("op_2241"), val = tensor([1, 0, 2])]; + tensor input_39_begin_0 = const()[name = string("input_39_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_39_end_0 = const()[name = string("input_39_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_39_end_mask_0 = const()[name = string("input_39_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_39_squeeze_mask_0 = const()[name = string("input_39_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_39_cast_fp16 = slice_by_index(begin = input_39_begin_0, end = input_39_end_0, end_mask = input_39_end_mask_0, squeeze_mask = input_39_squeeze_mask_0, x = z_5_cast_fp16)[name = string("input_39_cast_fp16")]; + fp16 var_1563_promoted_1_to_fp16 = const()[name = string("op_1563_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_69_cast_fp16 = transpose(perm = var_2241, x = var_2240_cast_fp16)[name = string("transpose_32")]; + tensor var_2245_cast_fp16 = pow(x = x_69_cast_fp16, y = var_1563_promoted_1_to_fp16)[name = string("op_2245_cast_fp16")]; + tensor var_2247_axes_0 = const()[name = string("op_2247_axes_0"), val = tensor([-1])]; + bool var_2247_keep_dims_0 = const()[name = string("op_2247_keep_dims_0"), val = bool(true)]; + tensor var_2247_cast_fp16 = reduce_mean(axes = var_2247_axes_0, keep_dims = var_2247_keep_dims_0, x = var_2245_cast_fp16)[name = string("op_2247_cast_fp16")]; + fp16 var_2248_to_fp16 = const()[name = string("op_2248_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2249_cast_fp16 = add(x = var_2247_cast_fp16, y = var_2248_to_fp16)[name = string("op_2249_cast_fp16")]; + fp32 var_2250_epsilon_0 = const()[name = string("op_2250_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2250_cast_fp16 = rsqrt(epsilon = var_2250_epsilon_0, x = var_2249_cast_fp16)[name = string("op_2250_cast_fp16")]; + tensor x_71_cast_fp16 = mul(x = x_69_cast_fp16, y = var_2250_cast_fp16)[name = string("x_71_cast_fp16")]; + tensor layers_2_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_2_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518799552)))]; + tensor var_2252_cast_fp16 = mul(x = layers_2_linear_attn_norm_weight_to_fp16, y = x_71_cast_fp16)[name = string("op_2252_cast_fp16")]; + tensor var_2253_cast_fp16 = silu(x = input_39_cast_fp16)[name = string("op_2253_cast_fp16")]; + tensor core_5_cast_fp16 = mul(x = var_2252_cast_fp16, y = var_2253_cast_fp16)[name = string("core_5_cast_fp16")]; + tensor var_2255 = const()[name = string("op_2255"), val = tensor([1, 1024, 4096])]; + tensor input_41_cast_fp16 = reshape(shape = var_2255, x = core_5_cast_fp16)[name = string("input_41_cast_fp16")]; + tensor layers_2_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_2_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(518799872)))]; + tensor linear_20_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_2_linear_attn_out_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_20_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_51_cast_fp16, y = linear_20_cast_fp16)[name = string("x_73_cast_fp16")]; + fp16 var_1563_promoted_2_to_fp16 = const()[name = string("op_1563_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2261_cast_fp16 = pow(x = x_73_cast_fp16, y = var_1563_promoted_2_to_fp16)[name = string("op_2261_cast_fp16")]; + tensor var_2263_axes_0 = const()[name = string("op_2263_axes_0"), val = tensor([-1])]; + bool var_2263_keep_dims_0 = const()[name = string("op_2263_keep_dims_0"), val = bool(true)]; + tensor var_2263_cast_fp16 = reduce_mean(axes = var_2263_axes_0, keep_dims = var_2263_keep_dims_0, x = var_2261_cast_fp16)[name = string("op_2263_cast_fp16")]; + fp16 var_2264_to_fp16 = const()[name = string("op_2264_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2265_cast_fp16 = add(x = var_2263_cast_fp16, y = var_2264_to_fp16)[name = string("op_2265_cast_fp16")]; + fp32 var_2266_epsilon_0 = const()[name = string("op_2266_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2266_cast_fp16 = rsqrt(epsilon = var_2266_epsilon_0, x = var_2265_cast_fp16)[name = string("op_2266_cast_fp16")]; + tensor x_75_cast_fp16 = mul(x = x_73_cast_fp16, y = var_2266_cast_fp16)[name = string("x_75_cast_fp16")]; + tensor var_2269_to_fp16 = const()[name = string("op_2269_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539771456)))]; + tensor input_43_cast_fp16 = mul(x = x_75_cast_fp16, y = var_2269_to_fp16)[name = string("input_43_cast_fp16")]; + tensor layers_2_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539776640)))]; + tensor linear_21_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_2_mlp_gate_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("linear_21_cast_fp16")]; + tensor var_2276_cast_fp16 = silu(x = linear_21_cast_fp16)[name = string("op_2276_cast_fp16")]; + tensor layers_2_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(586962624)))]; + tensor linear_22_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_2_mlp_up_proj_weight_to_fp16, x = input_43_cast_fp16)[name = string("linear_22_cast_fp16")]; + tensor input_47_cast_fp16 = mul(x = var_2276_cast_fp16, y = linear_22_cast_fp16)[name = string("input_47_cast_fp16")]; + tensor layers_2_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(634148608)))]; + tensor linear_23_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_2_mlp_down_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_23_cast_fp16")]; + tensor x_77_cast_fp16 = add(x = x_73_cast_fp16, y = linear_23_cast_fp16)[name = string("x_77_cast_fp16")]; + int32 var_2296 = const()[name = string("op_2296"), val = int32(-1)]; + fp16 var_2295_promoted_to_fp16 = const()[name = string("op_2295_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2305_cast_fp16 = pow(x = x_77_cast_fp16, y = var_2295_promoted_to_fp16)[name = string("op_2305_cast_fp16")]; + tensor var_2307_axes_0 = const()[name = string("op_2307_axes_0"), val = tensor([-1])]; + bool var_2307_keep_dims_0 = const()[name = string("op_2307_keep_dims_0"), val = bool(true)]; + tensor var_2307_cast_fp16 = reduce_mean(axes = var_2307_axes_0, keep_dims = var_2307_keep_dims_0, x = var_2305_cast_fp16)[name = string("op_2307_cast_fp16")]; + fp16 var_2308_to_fp16 = const()[name = string("op_2308_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2309_cast_fp16 = add(x = var_2307_cast_fp16, y = var_2308_to_fp16)[name = string("op_2309_cast_fp16")]; + fp32 var_2310_epsilon_0 = const()[name = string("op_2310_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2310_cast_fp16 = rsqrt(epsilon = var_2310_epsilon_0, x = var_2309_cast_fp16)[name = string("op_2310_cast_fp16")]; + tensor x_79_cast_fp16 = mul(x = x_77_cast_fp16, y = var_2310_cast_fp16)[name = string("x_79_cast_fp16")]; + tensor var_2313_to_fp16 = const()[name = string("op_2313_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(681334592)))]; + tensor x_81_cast_fp16 = mul(x = x_79_cast_fp16, y = var_2313_to_fp16)[name = string("x_81_cast_fp16")]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(681339776)))]; + tensor linear_24_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = x_81_cast_fp16)[name = string("linear_24_cast_fp16")]; + tensor var_2324 = const()[name = string("op_2324"), val = tensor([1, 1024, 16, 512])]; + tensor qg_1_cast_fp16 = reshape(shape = var_2324, x = linear_24_cast_fp16)[name = string("qg_1_cast_fp16")]; + tensor x_83_begin_0 = const()[name = string("x_83_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_83_end_0 = const()[name = string("x_83_end_0"), val = tensor([1, 1024, 16, 256])]; + tensor x_83_end_mask_0 = const()[name = string("x_83_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_83_cast_fp16 = slice_by_index(begin = x_83_begin_0, end = x_83_end_0, end_mask = x_83_end_mask_0, x = qg_1_cast_fp16)[name = string("x_83_cast_fp16")]; + tensor gate_1_begin_0 = const()[name = string("gate_1_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor gate_1_end_0 = const()[name = string("gate_1_end_0"), val = tensor([1, 1024, 16, 512])]; + tensor gate_1_end_mask_0 = const()[name = string("gate_1_end_mask_0"), val = tensor([true, true, true, true])]; + tensor gate_1_cast_fp16 = slice_by_index(begin = gate_1_begin_0, end = gate_1_end_0, end_mask = gate_1_end_mask_0, x = qg_1_cast_fp16)[name = string("gate_1_cast_fp16")]; + tensor var_2328 = const()[name = string("op_2328"), val = tensor([1, 1024, 4096])]; + tensor gate_3_cast_fp16 = reshape(shape = var_2328, x = gate_1_cast_fp16)[name = string("gate_3_cast_fp16")]; + fp16 var_2295_promoted_1_to_fp16 = const()[name = string("op_2295_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2331_cast_fp16 = pow(x = x_83_cast_fp16, y = var_2295_promoted_1_to_fp16)[name = string("op_2331_cast_fp16")]; + tensor var_2333_axes_0 = const()[name = string("op_2333_axes_0"), val = tensor([-1])]; + bool var_2333_keep_dims_0 = const()[name = string("op_2333_keep_dims_0"), val = bool(true)]; + tensor var_2333_cast_fp16 = reduce_mean(axes = var_2333_axes_0, keep_dims = var_2333_keep_dims_0, x = var_2331_cast_fp16)[name = string("op_2333_cast_fp16")]; + fp16 var_2334_to_fp16 = const()[name = string("op_2334_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2335_cast_fp16 = add(x = var_2333_cast_fp16, y = var_2334_to_fp16)[name = string("op_2335_cast_fp16")]; + fp32 var_2336_epsilon_0 = const()[name = string("op_2336_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2336_cast_fp16 = rsqrt(epsilon = var_2336_epsilon_0, x = var_2335_cast_fp16)[name = string("op_2336_cast_fp16")]; + tensor x_85_cast_fp16 = mul(x = x_83_cast_fp16, y = var_2336_cast_fp16)[name = string("x_85_cast_fp16")]; + tensor var_2339_to_fp16 = const()[name = string("op_2339_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723282880)))]; + tensor var_2340_cast_fp16 = mul(x = x_85_cast_fp16, y = var_2339_to_fp16)[name = string("op_2340_cast_fp16")]; + tensor q_31_perm_0 = const()[name = string("q_31_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723283456)))]; + tensor linear_25_bias_0_to_fp16 = const()[name = string("linear_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(728526400)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_25_bias_0_to_fp16, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = x_81_cast_fp16)[name = string("linear_25_cast_fp16")]; + tensor var_2344 = const()[name = string("op_2344"), val = tensor([1, 1024, 4, 256])]; + tensor x_87_cast_fp16 = reshape(shape = var_2344, x = linear_25_cast_fp16)[name = string("x_87_cast_fp16")]; + fp16 var_2295_promoted_2_to_fp16 = const()[name = string("op_2295_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2347_cast_fp16 = pow(x = x_87_cast_fp16, y = var_2295_promoted_2_to_fp16)[name = string("op_2347_cast_fp16")]; + tensor var_2349_axes_0 = const()[name = string("op_2349_axes_0"), val = tensor([-1])]; + bool var_2349_keep_dims_0 = const()[name = string("op_2349_keep_dims_0"), val = bool(true)]; + tensor var_2349_cast_fp16 = reduce_mean(axes = var_2349_axes_0, keep_dims = var_2349_keep_dims_0, x = var_2347_cast_fp16)[name = string("op_2349_cast_fp16")]; + fp16 var_2350_to_fp16 = const()[name = string("op_2350_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2351_cast_fp16 = add(x = var_2349_cast_fp16, y = var_2350_to_fp16)[name = string("op_2351_cast_fp16")]; + fp32 var_2352_epsilon_0 = const()[name = string("op_2352_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2352_cast_fp16 = rsqrt(epsilon = var_2352_epsilon_0, x = var_2351_cast_fp16)[name = string("op_2352_cast_fp16")]; + tensor x_89_cast_fp16 = mul(x = x_87_cast_fp16, y = var_2352_cast_fp16)[name = string("x_89_cast_fp16")]; + tensor var_2355_to_fp16 = const()[name = string("op_2355_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(728528512)))]; + tensor var_2356_cast_fp16 = mul(x = x_89_cast_fp16, y = var_2355_to_fp16)[name = string("op_2356_cast_fp16")]; + tensor k_25_perm_0 = const()[name = string("k_25_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(728529088)))]; + tensor linear_26_cast_fp16 = linear(bias = linear_25_bias_0_to_fp16, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = x_81_cast_fp16)[name = string("linear_26_cast_fp16")]; + tensor var_2360 = const()[name = string("op_2360"), val = tensor([1, 1024, 4, 256])]; + tensor var_2361_cast_fp16 = reshape(shape = var_2360, x = linear_26_cast_fp16)[name = string("op_2361_cast_fp16")]; + tensor v_13_perm_0 = const()[name = string("v_13_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_2363_axes_0 = const()[name = string("op_2363_axes_0"), val = tensor([0])]; + tensor var_2363_cast_fp16 = expand_dims(axes = var_2363_axes_0, x = cos)[name = string("op_2363_cast_fp16")]; + tensor c_1_axes_0 = const()[name = string("c_1_axes_0"), val = tensor([1])]; + tensor c_1_cast_fp16 = expand_dims(axes = c_1_axes_0, x = var_2363_cast_fp16)[name = string("c_1_cast_fp16")]; + tensor var_2365_axes_0 = const()[name = string("op_2365_axes_0"), val = tensor([0])]; + tensor var_2365_cast_fp16 = expand_dims(axes = var_2365_axes_0, x = sin)[name = string("op_2365_cast_fp16")]; + tensor s_1_axes_0 = const()[name = string("s_1_axes_0"), val = tensor([1])]; + tensor s_1_cast_fp16 = expand_dims(axes = s_1_axes_0, x = var_2365_cast_fp16)[name = string("s_1_cast_fp16")]; + tensor var_2367_begin_0 = const()[name = string("op_2367_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2367_end_0 = const()[name = string("op_2367_end_0"), val = tensor([1, 16, 1024, 64])]; + tensor var_2367_end_mask_0 = const()[name = string("op_2367_end_mask_0"), val = tensor([true, true, true, false])]; + tensor q_31_cast_fp16 = transpose(perm = q_31_perm_0, x = var_2340_cast_fp16)[name = string("transpose_31")]; + tensor var_2367_cast_fp16 = slice_by_index(begin = var_2367_begin_0, end = var_2367_end_0, end_mask = var_2367_end_mask_0, x = q_31_cast_fp16)[name = string("op_2367_cast_fp16")]; + tensor var_2368_cast_fp16 = mul(x = var_2367_cast_fp16, y = c_1_cast_fp16)[name = string("op_2368_cast_fp16")]; + tensor var_2375_begin_0 = const()[name = string("op_2375_begin_0"), val = tensor([0, 0, 0, 32])]; + tensor var_2375_end_0 = const()[name = string("op_2375_end_0"), val = tensor([1, 16, 1024, 64])]; + tensor var_2375_end_mask_0 = const()[name = string("op_2375_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2375_cast_fp16 = slice_by_index(begin = var_2375_begin_0, end = var_2375_end_0, end_mask = var_2375_end_mask_0, x = var_2367_cast_fp16)[name = string("op_2375_cast_fp16")]; + fp16 const_38_promoted_to_fp16 = const()[name = string("const_38_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_2376_cast_fp16 = mul(x = var_2375_cast_fp16, y = const_38_promoted_to_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2377_begin_0 = const()[name = string("op_2377_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2377_end_0 = const()[name = string("op_2377_end_0"), val = tensor([1, 16, 1024, 32])]; + tensor var_2377_end_mask_0 = const()[name = string("op_2377_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2377_cast_fp16 = slice_by_index(begin = var_2377_begin_0, end = var_2377_end_0, end_mask = var_2377_end_mask_0, x = var_2367_cast_fp16)[name = string("op_2377_cast_fp16")]; + bool var_2379_interleave_0 = const()[name = string("op_2379_interleave_0"), val = bool(false)]; + tensor var_2379_cast_fp16 = concat(axis = var_2296, interleave = var_2379_interleave_0, values = (var_2376_cast_fp16, var_2377_cast_fp16))[name = string("op_2379_cast_fp16")]; + tensor var_2380_cast_fp16 = mul(x = var_2379_cast_fp16, y = s_1_cast_fp16)[name = string("op_2380_cast_fp16")]; + tensor var_2381_cast_fp16 = add(x = var_2368_cast_fp16, y = var_2380_cast_fp16)[name = string("op_2381_cast_fp16")]; + tensor var_2382_begin_0 = const()[name = string("op_2382_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2382_end_0 = const()[name = string("op_2382_end_0"), val = tensor([1, 16, 1024, 256])]; + tensor var_2382_end_mask_0 = const()[name = string("op_2382_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2382_cast_fp16 = slice_by_index(begin = var_2382_begin_0, end = var_2382_end_0, end_mask = var_2382_end_mask_0, x = q_31_cast_fp16)[name = string("op_2382_cast_fp16")]; + bool q_33_interleave_0 = const()[name = string("q_33_interleave_0"), val = bool(false)]; + tensor q_33_cast_fp16 = concat(axis = var_2296, interleave = q_33_interleave_0, values = (var_2381_cast_fp16, var_2382_cast_fp16))[name = string("q_33_cast_fp16")]; + tensor var_2385_begin_0 = const()[name = string("op_2385_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2385_end_0 = const()[name = string("op_2385_end_0"), val = tensor([1, 4, 1024, 64])]; + tensor var_2385_end_mask_0 = const()[name = string("op_2385_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_25_cast_fp16 = transpose(perm = k_25_perm_0, x = var_2356_cast_fp16)[name = string("transpose_30")]; + tensor var_2385_cast_fp16 = slice_by_index(begin = var_2385_begin_0, end = var_2385_end_0, end_mask = var_2385_end_mask_0, x = k_25_cast_fp16)[name = string("op_2385_cast_fp16")]; + tensor var_2386_cast_fp16 = mul(x = var_2385_cast_fp16, y = c_1_cast_fp16)[name = string("op_2386_cast_fp16")]; + tensor var_2393_begin_0 = const()[name = string("op_2393_begin_0"), val = tensor([0, 0, 0, 32])]; + tensor var_2393_end_0 = const()[name = string("op_2393_end_0"), val = tensor([1, 4, 1024, 64])]; + tensor var_2393_end_mask_0 = const()[name = string("op_2393_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2393_cast_fp16 = slice_by_index(begin = var_2393_begin_0, end = var_2393_end_0, end_mask = var_2393_end_mask_0, x = var_2385_cast_fp16)[name = string("op_2393_cast_fp16")]; + fp16 const_40_promoted_to_fp16 = const()[name = string("const_40_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_2394_cast_fp16 = mul(x = var_2393_cast_fp16, y = const_40_promoted_to_fp16)[name = string("op_2394_cast_fp16")]; + tensor var_2395_begin_0 = const()[name = string("op_2395_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2395_end_0 = const()[name = string("op_2395_end_0"), val = tensor([1, 4, 1024, 32])]; + tensor var_2395_end_mask_0 = const()[name = string("op_2395_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2395_cast_fp16 = slice_by_index(begin = var_2395_begin_0, end = var_2395_end_0, end_mask = var_2395_end_mask_0, x = var_2385_cast_fp16)[name = string("op_2395_cast_fp16")]; + bool var_2397_interleave_0 = const()[name = string("op_2397_interleave_0"), val = bool(false)]; + tensor var_2397_cast_fp16 = concat(axis = var_2296, interleave = var_2397_interleave_0, values = (var_2394_cast_fp16, var_2395_cast_fp16))[name = string("op_2397_cast_fp16")]; + tensor var_2398_cast_fp16 = mul(x = var_2397_cast_fp16, y = s_1_cast_fp16)[name = string("op_2398_cast_fp16")]; + tensor var_2399_cast_fp16 = add(x = var_2386_cast_fp16, y = var_2398_cast_fp16)[name = string("op_2399_cast_fp16")]; + tensor var_2400_begin_0 = const()[name = string("op_2400_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2400_end_0 = const()[name = string("op_2400_end_0"), val = tensor([1, 4, 1024, 256])]; + tensor var_2400_end_mask_0 = const()[name = string("op_2400_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2400_cast_fp16 = slice_by_index(begin = var_2400_begin_0, end = var_2400_end_0, end_mask = var_2400_end_mask_0, x = k_25_cast_fp16)[name = string("op_2400_cast_fp16")]; + bool k_27_interleave_0 = const()[name = string("k_27_interleave_0"), val = bool(false)]; + tensor k_27_cast_fp16 = concat(axis = var_2296, interleave = k_27_interleave_0, values = (var_2399_cast_fp16, var_2400_cast_fp16))[name = string("k_27_cast_fp16")]; + tensor var_2405_axes_0 = const()[name = string("op_2405_axes_0"), val = tensor([2])]; + tensor var_2405_cast_fp16 = expand_dims(axes = var_2405_axes_0, x = k_27_cast_fp16)[name = string("op_2405_cast_fp16")]; + tensor var_2407_reps_0 = const()[name = string("op_2407_reps_0"), val = tensor([1, 1, 4, 1, 1])]; + tensor var_2407_cast_fp16 = tile(reps = var_2407_reps_0, x = var_2405_cast_fp16)[name = string("op_2407_cast_fp16")]; + tensor var_2408 = const()[name = string("op_2408"), val = tensor([1, 16, 1024, 256])]; + tensor k_29_cast_fp16 = reshape(shape = var_2408, x = var_2407_cast_fp16)[name = string("k_29_cast_fp16")]; + tensor var_2412_axes_0 = const()[name = string("op_2412_axes_0"), val = tensor([2])]; + tensor v_13_cast_fp16 = transpose(perm = v_13_perm_0, x = var_2361_cast_fp16)[name = string("transpose_29")]; + tensor var_2412_cast_fp16 = expand_dims(axes = var_2412_axes_0, x = v_13_cast_fp16)[name = string("op_2412_cast_fp16")]; + tensor var_2414_reps_0 = const()[name = string("op_2414_reps_0"), val = tensor([1, 1, 4, 1, 1])]; + tensor var_2414_cast_fp16 = tile(reps = var_2414_reps_0, x = var_2412_cast_fp16)[name = string("op_2414_cast_fp16")]; + tensor var_2415 = const()[name = string("op_2415"), val = tensor([1, 16, 1024, 256])]; + tensor v_15_cast_fp16 = reshape(shape = var_2415, x = var_2414_cast_fp16)[name = string("v_15_cast_fp16")]; + bool var_2418_transpose_x_1 = const()[name = string("op_2418_transpose_x_1"), val = bool(false)]; + bool var_2418_transpose_y_1 = const()[name = string("op_2418_transpose_y_1"), val = bool(true)]; + tensor var_2418_cast_fp16 = matmul(transpose_x = var_2418_transpose_x_1, transpose_y = var_2418_transpose_y_1, x = q_33_cast_fp16, y = k_29_cast_fp16)[name = string("op_2418_cast_fp16")]; + fp16 var_2419_to_fp16 = const()[name = string("op_2419_to_fp16"), val = fp16(0x1p-4)]; + tensor var_2420_cast_fp16 = mul(x = var_2418_cast_fp16, y = var_2419_to_fp16)[name = string("op_2420_cast_fp16")]; + tensor mask_to_fp16 = const()[name = string("mask_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(733772032)))]; + tensor scores_1_cast_fp16 = add(x = var_2420_cast_fp16, y = mask_to_fp16)[name = string("scores_1_cast_fp16")]; + tensor var_2422_cast_fp16 = softmax(axis = var_2296, x = scores_1_cast_fp16)[name = string("op_2422_cast_fp16")]; + bool out_1_transpose_x_0 = const()[name = string("out_1_transpose_x_0"), val = bool(false)]; + bool out_1_transpose_y_0 = const()[name = string("out_1_transpose_y_0"), val = bool(false)]; + tensor out_1_cast_fp16 = matmul(transpose_x = out_1_transpose_x_0, transpose_y = out_1_transpose_y_0, x = var_2422_cast_fp16, y = v_15_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor var_2424_perm_0 = const()[name = string("op_2424_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_2425 = const()[name = string("op_2425"), val = tensor([1, 1024, 4096])]; + tensor var_2424_cast_fp16 = transpose(perm = var_2424_perm_0, x = out_1_cast_fp16)[name = string("transpose_28")]; + tensor out_3_cast_fp16 = reshape(shape = var_2425, x = var_2424_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor var_2427_cast_fp16 = sigmoid(x = gate_3_cast_fp16)[name = string("op_2427_cast_fp16")]; + tensor input_49_cast_fp16 = mul(x = out_3_cast_fp16, y = var_2427_cast_fp16)[name = string("input_49_cast_fp16")]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(735869248)))]; + tensor linear_27_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_27_cast_fp16")]; + tensor x_95_cast_fp16 = add(x = x_77_cast_fp16, y = linear_27_cast_fp16)[name = string("x_95_cast_fp16")]; + fp16 var_2295_promoted_3_to_fp16 = const()[name = string("op_2295_promoted_3_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2433_cast_fp16 = pow(x = x_95_cast_fp16, y = var_2295_promoted_3_to_fp16)[name = string("op_2433_cast_fp16")]; + tensor var_2435_axes_0 = const()[name = string("op_2435_axes_0"), val = tensor([-1])]; + bool var_2435_keep_dims_0 = const()[name = string("op_2435_keep_dims_0"), val = bool(true)]; + tensor var_2435_cast_fp16 = reduce_mean(axes = var_2435_axes_0, keep_dims = var_2435_keep_dims_0, x = var_2433_cast_fp16)[name = string("op_2435_cast_fp16")]; + fp16 var_2436_to_fp16 = const()[name = string("op_2436_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2437_cast_fp16 = add(x = var_2435_cast_fp16, y = var_2436_to_fp16)[name = string("op_2437_cast_fp16")]; + fp32 var_2438_epsilon_0 = const()[name = string("op_2438_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2438_cast_fp16 = rsqrt(epsilon = var_2438_epsilon_0, x = var_2437_cast_fp16)[name = string("op_2438_cast_fp16")]; + tensor x_97_cast_fp16 = mul(x = x_95_cast_fp16, y = var_2438_cast_fp16)[name = string("x_97_cast_fp16")]; + tensor var_2441_to_fp16 = const()[name = string("op_2441_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(756840832)))]; + tensor input_51_cast_fp16 = mul(x = x_97_cast_fp16, y = var_2441_to_fp16)[name = string("input_51_cast_fp16")]; + tensor layers_3_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(756846016)))]; + tensor linear_28_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_3_mlp_gate_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_28_cast_fp16")]; + tensor var_2448_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("op_2448_cast_fp16")]; + tensor layers_3_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(804032000)))]; + tensor linear_29_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_3_mlp_up_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_29_cast_fp16")]; + tensor input_55_cast_fp16 = mul(x = var_2448_cast_fp16, y = linear_29_cast_fp16)[name = string("input_55_cast_fp16")]; + tensor layers_3_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(851217984)))]; + tensor linear_30_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_3_mlp_down_proj_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_30_cast_fp16")]; + tensor x_99_cast_fp16 = add(x = x_95_cast_fp16, y = linear_30_cast_fp16)[name = string("x_99_cast_fp16")]; + int32 var_2468 = const()[name = string("op_2468"), val = int32(-2)]; + int32 var_2489 = const()[name = string("op_2489"), val = int32(-1)]; + fp16 var_2488_promoted_to_fp16 = const()[name = string("op_2488_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_2498_cast_fp16 = pow(x = x_99_cast_fp16, y = var_2488_promoted_to_fp16)[name = string("op_2498_cast_fp16")]; + tensor var_2500_axes_0 = const()[name = string("op_2500_axes_0"), val = tensor([-1])]; + bool var_2500_keep_dims_0 = const()[name = string("op_2500_keep_dims_0"), val = bool(true)]; + tensor var_2500_cast_fp16 = reduce_mean(axes = var_2500_axes_0, keep_dims = var_2500_keep_dims_0, x = var_2498_cast_fp16)[name = string("op_2500_cast_fp16")]; + fp16 var_2501_to_fp16 = const()[name = string("op_2501_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2502_cast_fp16 = add(x = var_2500_cast_fp16, y = var_2501_to_fp16)[name = string("op_2502_cast_fp16")]; + fp32 var_2503_epsilon_0 = const()[name = string("op_2503_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2503_cast_fp16 = rsqrt(epsilon = var_2503_epsilon_0, x = var_2502_cast_fp16)[name = string("op_2503_cast_fp16")]; + tensor x_101_cast_fp16 = mul(x = x_99_cast_fp16, y = var_2503_cast_fp16)[name = string("x_101_cast_fp16")]; + tensor var_2506_to_fp16 = const()[name = string("op_2506_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(898403968)))]; + tensor x_103_cast_fp16 = mul(x = x_101_cast_fp16, y = var_2506_to_fp16)[name = string("x_103_cast_fp16")]; + tensor layers_4_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_4_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(898409152)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_linear_attn_in_proj_qkv_weight_to_fp16, x = x_103_cast_fp16)[name = string("linear_31_cast_fp16")]; + tensor input_57_perm_0 = const()[name = string("input_57_perm_0"), val = tensor([0, 2, 1])]; + tensor input_59_pad_0 = const()[name = string("input_59_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_59_mode_0 = const()[name = string("input_59_mode_0"), val = string("constant")]; + fp16 const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = fp16(0x0p+0)]; + tensor input_57_cast_fp16 = transpose(perm = input_57_perm_0, x = linear_31_cast_fp16)[name = string("transpose_27")]; + tensor input_59_cast_fp16 = pad(constant_val = const_42_to_fp16, mode = input_59_mode_0, pad = input_59_pad_0, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")]; + string input_61_pad_type_0 = const()[name = string("input_61_pad_type_0"), val = string("valid")]; + int32 input_61_groups_0 = const()[name = string("input_61_groups_0"), val = int32(8192)]; + tensor input_61_strides_0 = const()[name = string("input_61_strides_0"), val = tensor([1])]; + tensor input_61_pad_0 = const()[name = string("input_61_pad_0"), val = tensor([0, 0])]; + tensor input_61_dilations_0 = const()[name = string("input_61_dilations_0"), val = tensor([1])]; + tensor layers_4_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_4_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(940352256)))]; + tensor input_61_cast_fp16 = conv(dilations = input_61_dilations_0, groups = input_61_groups_0, pad = input_61_pad_0, pad_type = input_61_pad_type_0, strides = input_61_strides_0, weight = layers_4_linear_attn_conv1d_weight_to_fp16, x = input_59_cast_fp16)[name = string("input_61_cast_fp16")]; + tensor var_2535_cast_fp16 = silu(x = input_61_cast_fp16)[name = string("op_2535_cast_fp16")]; + tensor qkv_7_perm_0 = const()[name = string("qkv_7_perm_0"), val = tensor([0, 2, 1])]; + tensor var_2537_begin_0 = const()[name = string("op_2537_begin_0"), val = tensor([0, 0, 0])]; + tensor var_2537_end_0 = const()[name = string("op_2537_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_2537_end_mask_0 = const()[name = string("op_2537_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_7_cast_fp16 = transpose(perm = qkv_7_perm_0, x = var_2535_cast_fp16)[name = string("transpose_26")]; + tensor var_2537_cast_fp16 = slice_by_index(begin = var_2537_begin_0, end = var_2537_end_0, end_mask = var_2537_end_mask_0, x = qkv_7_cast_fp16)[name = string("op_2537_cast_fp16")]; + tensor var_2538 = const()[name = string("op_2538"), val = tensor([1, 1024, 16, 128])]; + tensor q_35_cast_fp16 = reshape(shape = var_2538, x = var_2537_cast_fp16)[name = string("q_35_cast_fp16")]; + tensor var_2540_begin_0 = const()[name = string("op_2540_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_2540_end_0 = const()[name = string("op_2540_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_2540_end_mask_0 = const()[name = string("op_2540_end_mask_0"), val = tensor([true, true, false])]; + tensor var_2540_cast_fp16 = slice_by_index(begin = var_2540_begin_0, end = var_2540_end_0, end_mask = var_2540_end_mask_0, x = qkv_7_cast_fp16)[name = string("op_2540_cast_fp16")]; + tensor var_2541 = const()[name = string("op_2541"), val = tensor([1, 1024, 16, 128])]; + tensor k_31_cast_fp16 = reshape(shape = var_2541, x = var_2540_cast_fp16)[name = string("k_31_cast_fp16")]; + tensor var_2543_begin_0 = const()[name = string("op_2543_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_2543_end_0 = const()[name = string("op_2543_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_2543_end_mask_0 = const()[name = string("op_2543_end_mask_0"), val = tensor([true, true, true])]; + tensor var_2543_cast_fp16 = slice_by_index(begin = var_2543_begin_0, end = var_2543_end_0, end_mask = var_2543_end_mask_0, x = qkv_7_cast_fp16)[name = string("op_2543_cast_fp16")]; + tensor var_2544 = const()[name = string("op_2544"), val = tensor([1, 1024, 32, 128])]; + tensor v_17_cast_fp16 = reshape(shape = var_2544, x = var_2543_cast_fp16)[name = string("v_17_cast_fp16")]; + tensor layers_4_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_4_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(940417856)))]; + tensor linear_32_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_4_linear_attn_in_proj_z_weight_to_fp16, x = x_103_cast_fp16)[name = string("linear_32_cast_fp16")]; + tensor var_2548 = const()[name = string("op_2548"), val = tensor([1, 1024, 32, 128])]; + tensor z_7_cast_fp16 = reshape(shape = var_2548, x = linear_32_cast_fp16)[name = string("z_7_cast_fp16")]; + tensor layers_4_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_4_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961389440)))]; + tensor linear_33_bias_0_to_fp16 = const()[name = string("linear_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961553344)))]; + tensor linear_33_cast_fp16 = linear(bias = linear_33_bias_0_to_fp16, weight = layers_4_linear_attn_in_proj_b_weight_to_fp16, x = x_103_cast_fp16)[name = string("linear_33_cast_fp16")]; + tensor beta_13_cast_fp16 = sigmoid(x = linear_33_cast_fp16)[name = string("beta_13_cast_fp16")]; + tensor var_2557_weight_0_to_fp16 = const()[name = string("op_2557_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961553472)))]; + tensor var_2557_bias_0_to_fp16 = const()[name = string("op_2557_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961717376)))]; + tensor var_2557_cast_fp16 = linear(bias = var_2557_bias_0_to_fp16, weight = var_2557_weight_0_to_fp16, x = x_103_cast_fp16)[name = string("op_2557_cast_fp16")]; + tensor var_2558_cast_fp16 = softplus(x = var_2557_cast_fp16)[name = string("op_2558_cast_fp16")]; + tensor var_2554_to_fp16 = const()[name = string("op_2554_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961717504)))]; + tensor g_13_cast_fp16 = mul(x = var_2554_to_fp16, y = var_2558_cast_fp16)[name = string("g_13_cast_fp16")]; + tensor var_2560_cast_fp16 = mul(x = q_35_cast_fp16, y = q_35_cast_fp16)[name = string("op_2560_cast_fp16")]; + tensor var_2562_axes_0 = const()[name = string("op_2562_axes_0"), val = tensor([-1])]; + bool var_2562_keep_dims_0 = const()[name = string("op_2562_keep_dims_0"), val = bool(true)]; + tensor var_2562_cast_fp16 = reduce_sum(axes = var_2562_axes_0, keep_dims = var_2562_keep_dims_0, x = var_2560_cast_fp16)[name = string("op_2562_cast_fp16")]; + fp16 var_2563_to_fp16 = const()[name = string("op_2563_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2564_cast_fp16 = add(x = var_2562_cast_fp16, y = var_2563_to_fp16)[name = string("op_2564_cast_fp16")]; + fp32 var_2565_epsilon_0 = const()[name = string("op_2565_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2565_cast_fp16 = rsqrt(epsilon = var_2565_epsilon_0, x = var_2564_cast_fp16)[name = string("op_2565_cast_fp16")]; + tensor q_37_cast_fp16 = mul(x = q_35_cast_fp16, y = var_2565_cast_fp16)[name = string("q_37_cast_fp16")]; + tensor var_2567_cast_fp16 = mul(x = k_31_cast_fp16, y = k_31_cast_fp16)[name = string("op_2567_cast_fp16")]; + tensor var_2569_axes_0 = const()[name = string("op_2569_axes_0"), val = tensor([-1])]; + bool var_2569_keep_dims_0 = const()[name = string("op_2569_keep_dims_0"), val = bool(true)]; + tensor var_2569_cast_fp16 = reduce_sum(axes = var_2569_axes_0, keep_dims = var_2569_keep_dims_0, x = var_2567_cast_fp16)[name = string("op_2569_cast_fp16")]; + fp16 var_2570_to_fp16 = const()[name = string("op_2570_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_2571_cast_fp16 = add(x = var_2569_cast_fp16, y = var_2570_to_fp16)[name = string("op_2571_cast_fp16")]; + fp32 var_2572_epsilon_0 = const()[name = string("op_2572_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_2572_cast_fp16 = rsqrt(epsilon = var_2572_epsilon_0, x = var_2571_cast_fp16)[name = string("op_2572_cast_fp16")]; + tensor k_33_cast_fp16 = mul(x = k_31_cast_fp16, y = var_2572_cast_fp16)[name = string("k_33_cast_fp16")]; + tensor var_2577_axes_0 = const()[name = string("op_2577_axes_0"), val = tensor([3])]; + tensor var_2577_cast_fp16 = expand_dims(axes = var_2577_axes_0, x = q_37_cast_fp16)[name = string("op_2577_cast_fp16")]; + tensor var_2579_reps_0 = const()[name = string("op_2579_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_2579_cast_fp16 = tile(reps = var_2579_reps_0, x = var_2577_cast_fp16)[name = string("op_2579_cast_fp16")]; + tensor var_2580 = const()[name = string("op_2580"), val = tensor([1, 1024, 32, 128])]; + tensor q_39_cast_fp16 = reshape(shape = var_2580, x = var_2579_cast_fp16)[name = string("q_39_cast_fp16")]; + tensor var_2585_axes_0 = const()[name = string("op_2585_axes_0"), val = tensor([3])]; + tensor var_2585_cast_fp16 = expand_dims(axes = var_2585_axes_0, x = k_33_cast_fp16)[name = string("op_2585_cast_fp16")]; + tensor var_2587_reps_0 = const()[name = string("op_2587_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_2587_cast_fp16 = tile(reps = var_2587_reps_0, x = var_2585_cast_fp16)[name = string("op_2587_cast_fp16")]; + tensor var_2588 = const()[name = string("op_2588"), val = tensor([1, 1024, 32, 128])]; + tensor k_35_cast_fp16 = reshape(shape = var_2588, x = var_2587_cast_fp16)[name = string("k_35_cast_fp16")]; + fp16 var_2590_to_fp16 = const()[name = string("op_2590_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_41_cast_fp16 = mul(x = q_39_cast_fp16, y = var_2590_to_fp16)[name = string("q_41_cast_fp16")]; + tensor var_2592_begin_0 = const()[name = string("op_2592_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2592_end_0 = const()[name = string("op_2592_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_2592_end_mask_0 = const()[name = string("op_2592_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_2592_squeeze_mask_0 = const()[name = string("op_2592_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_2592_cast_fp16 = slice_by_index(begin = var_2592_begin_0, end = var_2592_end_0, end_mask = var_2592_end_mask_0, squeeze_mask = var_2592_squeeze_mask_0, x = q_41_cast_fp16)[name = string("op_2592_cast_fp16")]; + tensor var_2593 = const()[name = string("op_2593"), val = tensor([1, 0, 2])]; + tensor var_2595 = const()[name = string("op_2595"), val = tensor([32, 16, 64, 128])]; + tensor var_2594_cast_fp16 = transpose(perm = var_2593, x = var_2592_cast_fp16)[name = string("transpose_25")]; + tensor q_43_cast_fp16 = reshape(shape = var_2595, x = var_2594_cast_fp16)[name = string("q_43_cast_fp16")]; + tensor var_2597_begin_0 = const()[name = string("op_2597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2597_end_0 = const()[name = string("op_2597_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_2597_end_mask_0 = const()[name = string("op_2597_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_2597_squeeze_mask_0 = const()[name = string("op_2597_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_2597_cast_fp16 = slice_by_index(begin = var_2597_begin_0, end = var_2597_end_0, end_mask = var_2597_end_mask_0, squeeze_mask = var_2597_squeeze_mask_0, x = k_35_cast_fp16)[name = string("op_2597_cast_fp16")]; + tensor var_2598 = const()[name = string("op_2598"), val = tensor([1, 0, 2])]; + tensor var_2600 = const()[name = string("op_2600"), val = tensor([32, 16, 64, 128])]; + tensor var_2599_cast_fp16 = transpose(perm = var_2598, x = var_2597_cast_fp16)[name = string("transpose_24")]; + tensor k_37_cast_fp16 = reshape(shape = var_2600, x = var_2599_cast_fp16)[name = string("k_37_cast_fp16")]; + tensor var_2602_begin_0 = const()[name = string("op_2602_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2602_end_0 = const()[name = string("op_2602_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_2602_end_mask_0 = const()[name = string("op_2602_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_2602_squeeze_mask_0 = const()[name = string("op_2602_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_2602_cast_fp16 = slice_by_index(begin = var_2602_begin_0, end = var_2602_end_0, end_mask = var_2602_end_mask_0, squeeze_mask = var_2602_squeeze_mask_0, x = v_17_cast_fp16)[name = string("op_2602_cast_fp16")]; + tensor var_2603 = const()[name = string("op_2603"), val = tensor([1, 0, 2])]; + tensor var_2605 = const()[name = string("op_2605"), val = tensor([32, 16, 64, 128])]; + tensor var_2604_cast_fp16 = transpose(perm = var_2603, x = var_2602_cast_fp16)[name = string("transpose_23")]; + tensor v_19_cast_fp16 = reshape(shape = var_2605, x = var_2604_cast_fp16)[name = string("v_19_cast_fp16")]; + tensor var_2607_begin_0 = const()[name = string("op_2607_begin_0"), val = tensor([0, 0, 0])]; + tensor var_2607_end_0 = const()[name = string("op_2607_end_0"), val = tensor([1, 1024, 32])]; + tensor var_2607_end_mask_0 = const()[name = string("op_2607_end_mask_0"), val = tensor([false, true, true])]; + tensor var_2607_squeeze_mask_0 = const()[name = string("op_2607_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_2607_cast_fp16 = slice_by_index(begin = var_2607_begin_0, end = var_2607_end_0, end_mask = var_2607_end_mask_0, squeeze_mask = var_2607_squeeze_mask_0, x = beta_13_cast_fp16)[name = string("op_2607_cast_fp16")]; + tensor var_2608_perm_0 = const()[name = string("op_2608_perm_0"), val = tensor([1, 0])]; + tensor var_2609 = const()[name = string("op_2609"), val = tensor([32, 16, 64, 1])]; + tensor var_2608_cast_fp16 = transpose(perm = var_2608_perm_0, x = var_2607_cast_fp16)[name = string("transpose_22")]; + tensor beta_15_cast_fp16 = reshape(shape = var_2609, x = var_2608_cast_fp16)[name = string("beta_15_cast_fp16")]; + tensor var_2611_begin_0 = const()[name = string("op_2611_begin_0"), val = tensor([0, 0, 0])]; + tensor var_2611_end_0 = const()[name = string("op_2611_end_0"), val = tensor([1, 1024, 32])]; + tensor var_2611_end_mask_0 = const()[name = string("op_2611_end_mask_0"), val = tensor([false, true, true])]; + tensor var_2611_squeeze_mask_0 = const()[name = string("op_2611_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_2611_cast_fp16 = slice_by_index(begin = var_2611_begin_0, end = var_2611_end_0, end_mask = var_2611_end_mask_0, squeeze_mask = var_2611_squeeze_mask_0, x = g_13_cast_fp16)[name = string("op_2611_cast_fp16")]; + tensor var_2612_perm_0 = const()[name = string("op_2612_perm_0"), val = tensor([1, 0])]; + tensor var_2613 = const()[name = string("op_2613"), val = tensor([32, 16, 64])]; + tensor var_2612_cast_fp16 = transpose(perm = var_2612_perm_0, x = var_2611_cast_fp16)[name = string("transpose_21")]; + tensor g_15_cast_fp16 = reshape(shape = var_2613, x = var_2612_cast_fp16)[name = string("g_15_cast_fp16")]; + tensor g_col_7_axes_0 = const()[name = string("g_col_7_axes_0"), val = tensor([-1])]; + tensor g_col_7_cast_fp16 = expand_dims(axes = g_col_7_axes_0, x = g_15_cast_fp16)[name = string("g_col_7_cast_fp16")]; + bool var_2616_transpose_x_0 = const()[name = string("op_2616_transpose_x_0"), val = bool(false)]; + bool var_2616_transpose_y_0 = const()[name = string("op_2616_transpose_y_0"), val = bool(false)]; + tensor var_2616_cast_fp16 = matmul(transpose_x = var_2616_transpose_x_0, transpose_y = var_2616_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_7_cast_fp16)[name = string("op_2616_cast_fp16")]; + tensor cum_7_axes_0 = const()[name = string("cum_7_axes_0"), val = tensor([-1])]; + tensor cum_7_cast_fp16 = squeeze(axes = cum_7_axes_0, x = var_2616_cast_fp16)[name = string("cum_7_cast_fp16")]; + bool var_2618_transpose_x_0 = const()[name = string("op_2618_transpose_x_0"), val = bool(false)]; + bool var_2618_transpose_y_0 = const()[name = string("op_2618_transpose_y_0"), val = bool(false)]; + tensor var_2618_cast_fp16 = matmul(transpose_x = var_2618_transpose_x_0, transpose_y = var_2618_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_7_cast_fp16)[name = string("op_2618_cast_fp16")]; + tensor to_end_7_axes_0 = const()[name = string("to_end_7_axes_0"), val = tensor([-1])]; + tensor to_end_7_cast_fp16 = squeeze(axes = to_end_7_axes_0, x = var_2618_cast_fp16)[name = string("to_end_7_cast_fp16")]; + bool var_2620_transpose_x_0 = const()[name = string("op_2620_transpose_x_0"), val = bool(false)]; + bool var_2620_transpose_y_0 = const()[name = string("op_2620_transpose_y_0"), val = bool(false)]; + tensor var_2620_cast_fp16 = matmul(transpose_x = var_2620_transpose_x_0, transpose_y = var_2620_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_7_cast_fp16)[name = string("op_2620_cast_fp16")]; + tensor var_2621 = const()[name = string("op_2621"), val = tensor([32, 16, 64, 64])]; + tensor pair_43_cast_fp16 = reshape(shape = var_2621, x = var_2620_cast_fp16)[name = string("pair_43_cast_fp16")]; + tensor var_2623_cast_fp16 = exp(x = pair_43_cast_fp16)[name = string("op_2623_cast_fp16")]; + tensor pair_decay_7_cast_fp16 = mul(x = var_2623_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_7_cast_fp16")]; + tensor k_beta_7_cast_fp16 = mul(x = k_37_cast_fp16, y = beta_15_cast_fp16)[name = string("k_beta_7_cast_fp16")]; + tensor v_beta_7_cast_fp16 = mul(x = v_19_cast_fp16, y = beta_15_cast_fp16)[name = string("v_beta_7_cast_fp16")]; + bool var_2628_transpose_x_1 = const()[name = string("op_2628_transpose_x_1"), val = bool(false)]; + bool var_2628_transpose_y_1 = const()[name = string("op_2628_transpose_y_1"), val = bool(true)]; + tensor var_2628_cast_fp16 = matmul(transpose_x = var_2628_transpose_x_1, transpose_y = var_2628_transpose_y_1, x = k_beta_7_cast_fp16, y = k_37_cast_fp16)[name = string("op_2628_cast_fp16")]; + tensor kkt_7_cast_fp16 = mul(x = var_2628_cast_fp16, y = pair_decay_7_cast_fp16)[name = string("kkt_7_cast_fp16")]; + bool var_2631_transpose_x_1 = const()[name = string("op_2631_transpose_x_1"), val = bool(false)]; + bool var_2631_transpose_y_1 = const()[name = string("op_2631_transpose_y_1"), val = bool(true)]; + tensor var_2631_cast_fp16 = matmul(transpose_x = var_2631_transpose_x_1, transpose_y = var_2631_transpose_y_1, x = q_43_cast_fp16, y = k_37_cast_fp16)[name = string("op_2631_cast_fp16")]; + tensor attn_intra_7_cast_fp16 = mul(x = var_2631_cast_fp16, y = pair_decay_7_cast_fp16)[name = string("attn_intra_7_cast_fp16")]; + tensor var_2633_cast_fp16 = mul(x = kkt_7_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_2633_cast_fp16")]; + tensor t_13_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_2633_cast_fp16)[name = string("t_13_cast_fp16")]; + tensor var_2637 = const()[name = string("op_2637"), val = tensor([-1, 64, 64])]; + tensor t_15_cast_fp16 = reshape(shape = var_2637, x = t_13_cast_fp16)[name = string("t_15_cast_fp16")]; + tensor var_2642 = const()[name = string("op_2642"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_2643_cast_fp16 = reshape(shape = var_2642, x = t_15_cast_fp16)[name = string("op_2643_cast_fp16")]; + tensor var_2648_cast_fp16 = mul(x = var_2643_cast_fp16, y = var_216_to_fp16)[name = string("op_2648_cast_fp16")]; + tensor blocks_37_axes_0 = const()[name = string("blocks_37_axes_0"), val = tensor([-2])]; + bool blocks_37_keep_dims_0 = const()[name = string("blocks_37_keep_dims_0"), val = bool(false)]; + tensor blocks_37_cast_fp16 = reduce_sum(axes = blocks_37_axes_0, keep_dims = blocks_37_keep_dims_0, x = var_2648_cast_fp16)[name = string("blocks_37_cast_fp16")]; + tensor var_2651_begin_0 = const()[name = string("op_2651_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2651_end_0 = const()[name = string("op_2651_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_2651_end_mask_0 = const()[name = string("op_2651_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2651_cast_fp16 = slice_by_index(begin = var_2651_begin_0, end = var_2651_end_0, end_mask = var_2651_end_mask_0, x = blocks_37_cast_fp16)[name = string("op_2651_cast_fp16")]; + tensor x_105_begin_0 = const()[name = string("x_105_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_105_end_0 = const()[name = string("x_105_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_105_end_mask_0 = const()[name = string("x_105_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_105_cast_fp16 = slice_by_index(begin = x_105_begin_0, end = x_105_end_0, end_mask = x_105_end_mask_0, x = var_2651_cast_fp16)[name = string("x_105_cast_fp16")]; + bool var_2661_transpose_x_0 = const()[name = string("op_2661_transpose_x_0"), val = bool(false)]; + bool var_2661_transpose_y_0 = const()[name = string("op_2661_transpose_y_0"), val = bool(false)]; + tensor var_2661_cast_fp16 = matmul(transpose_x = var_2661_transpose_x_0, transpose_y = var_2661_transpose_y_0, x = x_105_cast_fp16, y = var_224_to_fp16)[name = string("op_2661_cast_fp16")]; + bool var_2662_transpose_x_0 = const()[name = string("op_2662_transpose_x_0"), val = bool(false)]; + bool var_2662_transpose_y_0 = const()[name = string("op_2662_transpose_y_0"), val = bool(false)]; + tensor var_2662_cast_fp16 = matmul(transpose_x = var_2662_transpose_x_0, transpose_y = var_2662_transpose_y_0, x = var_224_to_fp16, y = var_2661_cast_fp16)[name = string("op_2662_cast_fp16")]; + fp16 const_47_promoted_to_fp16 = const()[name = string("const_47_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_37_cast_fp16 = mul(x = var_2662_cast_fp16, y = const_47_promoted_to_fp16)[name = string("lower_37_cast_fp16")]; + bool bottom_37_interleave_0 = const()[name = string("bottom_37_interleave_0"), val = bool(false)]; + tensor bottom_37_cast_fp16 = concat(axis = var_2489, interleave = bottom_37_interleave_0, values = (lower_37_cast_fp16, var_224_to_fp16))[name = string("bottom_37_cast_fp16")]; + bool inv_51_interleave_0 = const()[name = string("inv_51_interleave_0"), val = bool(false)]; + tensor inv_51_cast_fp16 = concat(axis = var_2468, interleave = inv_51_interleave_0, values = (top_1_to_fp16, bottom_37_cast_fp16))[name = string("inv_51_cast_fp16")]; + tensor var_2671 = const()[name = string("op_2671"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_2672_cast_fp16 = reshape(shape = var_2671, x = t_15_cast_fp16)[name = string("op_2672_cast_fp16")]; + tensor var_2677_cast_fp16 = mul(x = var_2672_cast_fp16, y = var_245_to_fp16)[name = string("op_2677_cast_fp16")]; + tensor blocks_39_axes_0 = const()[name = string("blocks_39_axes_0"), val = tensor([-2])]; + bool blocks_39_keep_dims_0 = const()[name = string("blocks_39_keep_dims_0"), val = bool(false)]; + tensor blocks_39_cast_fp16 = reduce_sum(axes = blocks_39_axes_0, keep_dims = blocks_39_keep_dims_0, x = var_2677_cast_fp16)[name = string("blocks_39_cast_fp16")]; + tensor var_2680_begin_0 = const()[name = string("op_2680_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_2680_end_0 = const()[name = string("op_2680_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_2680_end_mask_0 = const()[name = string("op_2680_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2680_cast_fp16 = slice_by_index(begin = var_2680_begin_0, end = var_2680_end_0, end_mask = var_2680_end_mask_0, x = blocks_39_cast_fp16)[name = string("op_2680_cast_fp16")]; + tensor x_107_begin_0 = const()[name = string("x_107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_107_end_0 = const()[name = string("x_107_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_107_end_mask_0 = const()[name = string("x_107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_107_cast_fp16 = slice_by_index(begin = x_107_begin_0, end = x_107_end_0, end_mask = x_107_end_mask_0, x = var_2680_cast_fp16)[name = string("x_107_cast_fp16")]; + tensor var_2682 = const()[name = string("op_2682"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_47_cast_fp16 = reshape(shape = var_2682, x = inv_51_cast_fp16)[name = string("pair_47_cast_fp16")]; + tensor var_2684_begin_0 = const()[name = string("op_2684_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_2684_end_0 = const()[name = string("op_2684_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_2684_end_mask_0 = const()[name = string("op_2684_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2684_squeeze_mask_0 = const()[name = string("op_2684_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2684_cast_fp16 = slice_by_index(begin = var_2684_begin_0, end = var_2684_end_0, end_mask = var_2684_end_mask_0, squeeze_mask = var_2684_squeeze_mask_0, x = pair_47_cast_fp16)[name = string("op_2684_cast_fp16")]; + tensor var_2687_begin_0 = const()[name = string("op_2687_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_2687_end_0 = const()[name = string("op_2687_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_2687_end_mask_0 = const()[name = string("op_2687_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2687_squeeze_mask_0 = const()[name = string("op_2687_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2687_cast_fp16 = slice_by_index(begin = var_2687_begin_0, end = var_2687_end_0, end_mask = var_2687_end_mask_0, squeeze_mask = var_2687_squeeze_mask_0, x = pair_47_cast_fp16)[name = string("op_2687_cast_fp16")]; + bool var_2690_transpose_x_0 = const()[name = string("op_2690_transpose_x_0"), val = bool(false)]; + bool var_2690_transpose_y_0 = const()[name = string("op_2690_transpose_y_0"), val = bool(false)]; + tensor var_2690_cast_fp16 = matmul(transpose_x = var_2690_transpose_x_0, transpose_y = var_2690_transpose_y_0, x = x_107_cast_fp16, y = var_2684_cast_fp16)[name = string("op_2690_cast_fp16")]; + bool var_2691_transpose_x_0 = const()[name = string("op_2691_transpose_x_0"), val = bool(false)]; + bool var_2691_transpose_y_0 = const()[name = string("op_2691_transpose_y_0"), val = bool(false)]; + tensor var_2691_cast_fp16 = matmul(transpose_x = var_2691_transpose_x_0, transpose_y = var_2691_transpose_y_0, x = var_2687_cast_fp16, y = var_2690_cast_fp16)[name = string("op_2691_cast_fp16")]; + fp16 const_48_promoted_to_fp16 = const()[name = string("const_48_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_39_cast_fp16 = mul(x = var_2691_cast_fp16, y = const_48_promoted_to_fp16)[name = string("lower_39_cast_fp16")]; + bool top_39_interleave_0 = const()[name = string("top_39_interleave_0"), val = bool(false)]; + tensor top_39_cast_fp16 = concat(axis = var_2489, interleave = top_39_interleave_0, values = (var_2684_cast_fp16, var_262_to_fp16))[name = string("top_39_cast_fp16")]; + bool bottom_39_interleave_0 = const()[name = string("bottom_39_interleave_0"), val = bool(false)]; + tensor bottom_39_cast_fp16 = concat(axis = var_2489, interleave = bottom_39_interleave_0, values = (lower_39_cast_fp16, var_2687_cast_fp16))[name = string("bottom_39_cast_fp16")]; + bool inv_53_interleave_0 = const()[name = string("inv_53_interleave_0"), val = bool(false)]; + tensor inv_53_cast_fp16 = concat(axis = var_2468, interleave = inv_53_interleave_0, values = (top_39_cast_fp16, bottom_39_cast_fp16))[name = string("inv_53_cast_fp16")]; + tensor var_2700 = const()[name = string("op_2700"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_2701_cast_fp16 = reshape(shape = var_2700, x = t_15_cast_fp16)[name = string("op_2701_cast_fp16")]; + tensor var_2705_to_fp16 = const()[name = string("op_2705_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961717632)))]; + tensor var_2706_cast_fp16 = mul(x = var_2701_cast_fp16, y = var_2705_to_fp16)[name = string("op_2706_cast_fp16")]; + tensor blocks_41_axes_0 = const()[name = string("blocks_41_axes_0"), val = tensor([-2])]; + bool blocks_41_keep_dims_0 = const()[name = string("blocks_41_keep_dims_0"), val = bool(false)]; + tensor blocks_41_cast_fp16 = reduce_sum(axes = blocks_41_axes_0, keep_dims = blocks_41_keep_dims_0, x = var_2706_cast_fp16)[name = string("blocks_41_cast_fp16")]; + tensor var_2709_begin_0 = const()[name = string("op_2709_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_2709_end_0 = const()[name = string("op_2709_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_2709_end_mask_0 = const()[name = string("op_2709_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2709_cast_fp16 = slice_by_index(begin = var_2709_begin_0, end = var_2709_end_0, end_mask = var_2709_end_mask_0, x = blocks_41_cast_fp16)[name = string("op_2709_cast_fp16")]; + tensor x_109_begin_0 = const()[name = string("x_109_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_109_end_0 = const()[name = string("x_109_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_109_end_mask_0 = const()[name = string("x_109_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_109_cast_fp16 = slice_by_index(begin = x_109_begin_0, end = x_109_end_0, end_mask = x_109_end_mask_0, x = var_2709_cast_fp16)[name = string("x_109_cast_fp16")]; + tensor var_2711 = const()[name = string("op_2711"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_49_cast_fp16 = reshape(shape = var_2711, x = inv_53_cast_fp16)[name = string("pair_49_cast_fp16")]; + tensor var_2713_begin_0 = const()[name = string("op_2713_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_2713_end_0 = const()[name = string("op_2713_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_2713_end_mask_0 = const()[name = string("op_2713_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2713_squeeze_mask_0 = const()[name = string("op_2713_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2713_cast_fp16 = slice_by_index(begin = var_2713_begin_0, end = var_2713_end_0, end_mask = var_2713_end_mask_0, squeeze_mask = var_2713_squeeze_mask_0, x = pair_49_cast_fp16)[name = string("op_2713_cast_fp16")]; + tensor var_2716_begin_0 = const()[name = string("op_2716_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_2716_end_0 = const()[name = string("op_2716_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_2716_end_mask_0 = const()[name = string("op_2716_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2716_squeeze_mask_0 = const()[name = string("op_2716_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2716_cast_fp16 = slice_by_index(begin = var_2716_begin_0, end = var_2716_end_0, end_mask = var_2716_end_mask_0, squeeze_mask = var_2716_squeeze_mask_0, x = pair_49_cast_fp16)[name = string("op_2716_cast_fp16")]; + bool var_2719_transpose_x_0 = const()[name = string("op_2719_transpose_x_0"), val = bool(false)]; + bool var_2719_transpose_y_0 = const()[name = string("op_2719_transpose_y_0"), val = bool(false)]; + tensor var_2719_cast_fp16 = matmul(transpose_x = var_2719_transpose_x_0, transpose_y = var_2719_transpose_y_0, x = x_109_cast_fp16, y = var_2713_cast_fp16)[name = string("op_2719_cast_fp16")]; + bool var_2720_transpose_x_0 = const()[name = string("op_2720_transpose_x_0"), val = bool(false)]; + bool var_2720_transpose_y_0 = const()[name = string("op_2720_transpose_y_0"), val = bool(false)]; + tensor var_2720_cast_fp16 = matmul(transpose_x = var_2720_transpose_x_0, transpose_y = var_2720_transpose_y_0, x = var_2716_cast_fp16, y = var_2719_cast_fp16)[name = string("op_2720_cast_fp16")]; + fp16 const_49_promoted_to_fp16 = const()[name = string("const_49_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_41_cast_fp16 = mul(x = var_2720_cast_fp16, y = const_49_promoted_to_fp16)[name = string("lower_41_cast_fp16")]; + bool top_41_interleave_0 = const()[name = string("top_41_interleave_0"), val = bool(false)]; + tensor top_41_cast_fp16 = concat(axis = var_2489, interleave = top_41_interleave_0, values = (var_2713_cast_fp16, var_291_to_fp16))[name = string("top_41_cast_fp16")]; + bool bottom_41_interleave_0 = const()[name = string("bottom_41_interleave_0"), val = bool(false)]; + tensor bottom_41_cast_fp16 = concat(axis = var_2489, interleave = bottom_41_interleave_0, values = (lower_41_cast_fp16, var_2716_cast_fp16))[name = string("bottom_41_cast_fp16")]; + bool inv_55_interleave_0 = const()[name = string("inv_55_interleave_0"), val = bool(false)]; + tensor inv_55_cast_fp16 = concat(axis = var_2468, interleave = inv_55_interleave_0, values = (top_41_cast_fp16, bottom_41_cast_fp16))[name = string("inv_55_cast_fp16")]; + tensor var_2729 = const()[name = string("op_2729"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_2730_cast_fp16 = reshape(shape = var_2729, x = t_15_cast_fp16)[name = string("op_2730_cast_fp16")]; + tensor var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961717824)))]; + tensor var_2735_cast_fp16 = mul(x = var_2730_cast_fp16, y = var_2734_to_fp16)[name = string("op_2735_cast_fp16")]; + tensor blocks_43_axes_0 = const()[name = string("blocks_43_axes_0"), val = tensor([-2])]; + bool blocks_43_keep_dims_0 = const()[name = string("blocks_43_keep_dims_0"), val = bool(false)]; + tensor blocks_43_cast_fp16 = reduce_sum(axes = blocks_43_axes_0, keep_dims = blocks_43_keep_dims_0, x = var_2735_cast_fp16)[name = string("blocks_43_cast_fp16")]; + tensor var_2738_begin_0 = const()[name = string("op_2738_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_2738_end_0 = const()[name = string("op_2738_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_2738_end_mask_0 = const()[name = string("op_2738_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2738_cast_fp16 = slice_by_index(begin = var_2738_begin_0, end = var_2738_end_0, end_mask = var_2738_end_mask_0, x = blocks_43_cast_fp16)[name = string("op_2738_cast_fp16")]; + tensor x_111_begin_0 = const()[name = string("x_111_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_111_end_0 = const()[name = string("x_111_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_111_end_mask_0 = const()[name = string("x_111_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_111_cast_fp16 = slice_by_index(begin = x_111_begin_0, end = x_111_end_0, end_mask = x_111_end_mask_0, x = var_2738_cast_fp16)[name = string("x_111_cast_fp16")]; + tensor var_2740 = const()[name = string("op_2740"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_51_cast_fp16 = reshape(shape = var_2740, x = inv_55_cast_fp16)[name = string("pair_51_cast_fp16")]; + tensor var_2742_begin_0 = const()[name = string("op_2742_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_2742_end_0 = const()[name = string("op_2742_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_2742_end_mask_0 = const()[name = string("op_2742_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2742_squeeze_mask_0 = const()[name = string("op_2742_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2742_cast_fp16 = slice_by_index(begin = var_2742_begin_0, end = var_2742_end_0, end_mask = var_2742_end_mask_0, squeeze_mask = var_2742_squeeze_mask_0, x = pair_51_cast_fp16)[name = string("op_2742_cast_fp16")]; + tensor var_2745_begin_0 = const()[name = string("op_2745_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_2745_end_0 = const()[name = string("op_2745_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_2745_end_mask_0 = const()[name = string("op_2745_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2745_squeeze_mask_0 = const()[name = string("op_2745_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2745_cast_fp16 = slice_by_index(begin = var_2745_begin_0, end = var_2745_end_0, end_mask = var_2745_end_mask_0, squeeze_mask = var_2745_squeeze_mask_0, x = pair_51_cast_fp16)[name = string("op_2745_cast_fp16")]; + bool var_2748_transpose_x_0 = const()[name = string("op_2748_transpose_x_0"), val = bool(false)]; + bool var_2748_transpose_y_0 = const()[name = string("op_2748_transpose_y_0"), val = bool(false)]; + tensor var_2748_cast_fp16 = matmul(transpose_x = var_2748_transpose_x_0, transpose_y = var_2748_transpose_y_0, x = x_111_cast_fp16, y = var_2742_cast_fp16)[name = string("op_2748_cast_fp16")]; + bool var_2749_transpose_x_0 = const()[name = string("op_2749_transpose_x_0"), val = bool(false)]; + bool var_2749_transpose_y_0 = const()[name = string("op_2749_transpose_y_0"), val = bool(false)]; + tensor var_2749_cast_fp16 = matmul(transpose_x = var_2749_transpose_x_0, transpose_y = var_2749_transpose_y_0, x = var_2745_cast_fp16, y = var_2748_cast_fp16)[name = string("op_2749_cast_fp16")]; + fp16 const_50_promoted_to_fp16 = const()[name = string("const_50_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_43_cast_fp16 = mul(x = var_2749_cast_fp16, y = const_50_promoted_to_fp16)[name = string("lower_43_cast_fp16")]; + bool top_43_interleave_0 = const()[name = string("top_43_interleave_0"), val = bool(false)]; + tensor top_43_cast_fp16 = concat(axis = var_2489, interleave = top_43_interleave_0, values = (var_2742_cast_fp16, var_320_to_fp16))[name = string("top_43_cast_fp16")]; + bool bottom_43_interleave_0 = const()[name = string("bottom_43_interleave_0"), val = bool(false)]; + tensor bottom_43_cast_fp16 = concat(axis = var_2489, interleave = bottom_43_interleave_0, values = (lower_43_cast_fp16, var_2745_cast_fp16))[name = string("bottom_43_cast_fp16")]; + bool inv_57_interleave_0 = const()[name = string("inv_57_interleave_0"), val = bool(false)]; + tensor inv_57_cast_fp16 = concat(axis = var_2468, interleave = inv_57_interleave_0, values = (top_43_cast_fp16, bottom_43_cast_fp16))[name = string("inv_57_cast_fp16")]; + tensor var_2758 = const()[name = string("op_2758"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_2759_cast_fp16 = reshape(shape = var_2758, x = t_15_cast_fp16)[name = string("op_2759_cast_fp16")]; + tensor var_2763_to_fp16 = const()[name = string("op_2763_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_2764_cast_fp16 = mul(x = var_2759_cast_fp16, y = var_2763_to_fp16)[name = string("op_2764_cast_fp16")]; + tensor blocks_45_axes_0 = const()[name = string("blocks_45_axes_0"), val = tensor([-2])]; + bool blocks_45_keep_dims_0 = const()[name = string("blocks_45_keep_dims_0"), val = bool(false)]; + tensor blocks_45_cast_fp16 = reduce_sum(axes = blocks_45_axes_0, keep_dims = blocks_45_keep_dims_0, x = var_2764_cast_fp16)[name = string("blocks_45_cast_fp16")]; + tensor var_2767_begin_0 = const()[name = string("op_2767_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_2767_end_0 = const()[name = string("op_2767_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_2767_end_mask_0 = const()[name = string("op_2767_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2767_cast_fp16 = slice_by_index(begin = var_2767_begin_0, end = var_2767_end_0, end_mask = var_2767_end_mask_0, x = blocks_45_cast_fp16)[name = string("op_2767_cast_fp16")]; + tensor x_113_begin_0 = const()[name = string("x_113_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_113_end_0 = const()[name = string("x_113_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_113_end_mask_0 = const()[name = string("x_113_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_113_cast_fp16 = slice_by_index(begin = x_113_begin_0, end = x_113_end_0, end_mask = x_113_end_mask_0, x = var_2767_cast_fp16)[name = string("x_113_cast_fp16")]; + tensor var_2769 = const()[name = string("op_2769"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_53_cast_fp16 = reshape(shape = var_2769, x = inv_57_cast_fp16)[name = string("pair_53_cast_fp16")]; + tensor var_2771_begin_0 = const()[name = string("op_2771_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_2771_end_0 = const()[name = string("op_2771_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_2771_end_mask_0 = const()[name = string("op_2771_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2771_squeeze_mask_0 = const()[name = string("op_2771_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2771_cast_fp16 = slice_by_index(begin = var_2771_begin_0, end = var_2771_end_0, end_mask = var_2771_end_mask_0, squeeze_mask = var_2771_squeeze_mask_0, x = pair_53_cast_fp16)[name = string("op_2771_cast_fp16")]; + tensor var_2774_begin_0 = const()[name = string("op_2774_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_2774_end_0 = const()[name = string("op_2774_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_2774_end_mask_0 = const()[name = string("op_2774_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2774_squeeze_mask_0 = const()[name = string("op_2774_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2774_cast_fp16 = slice_by_index(begin = var_2774_begin_0, end = var_2774_end_0, end_mask = var_2774_end_mask_0, squeeze_mask = var_2774_squeeze_mask_0, x = pair_53_cast_fp16)[name = string("op_2774_cast_fp16")]; + bool var_2777_transpose_x_0 = const()[name = string("op_2777_transpose_x_0"), val = bool(false)]; + bool var_2777_transpose_y_0 = const()[name = string("op_2777_transpose_y_0"), val = bool(false)]; + tensor var_2777_cast_fp16 = matmul(transpose_x = var_2777_transpose_x_0, transpose_y = var_2777_transpose_y_0, x = x_113_cast_fp16, y = var_2771_cast_fp16)[name = string("op_2777_cast_fp16")]; + bool var_2778_transpose_x_0 = const()[name = string("op_2778_transpose_x_0"), val = bool(false)]; + bool var_2778_transpose_y_0 = const()[name = string("op_2778_transpose_y_0"), val = bool(false)]; + tensor var_2778_cast_fp16 = matmul(transpose_x = var_2778_transpose_x_0, transpose_y = var_2778_transpose_y_0, x = var_2774_cast_fp16, y = var_2777_cast_fp16)[name = string("op_2778_cast_fp16")]; + fp16 const_51_promoted_to_fp16 = const()[name = string("const_51_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_45_cast_fp16 = mul(x = var_2778_cast_fp16, y = const_51_promoted_to_fp16)[name = string("lower_45_cast_fp16")]; + bool top_45_interleave_0 = const()[name = string("top_45_interleave_0"), val = bool(false)]; + tensor top_45_cast_fp16 = concat(axis = var_2489, interleave = top_45_interleave_0, values = (var_2771_cast_fp16, var_349_to_fp16))[name = string("top_45_cast_fp16")]; + bool bottom_45_interleave_0 = const()[name = string("bottom_45_interleave_0"), val = bool(false)]; + tensor bottom_45_cast_fp16 = concat(axis = var_2489, interleave = bottom_45_interleave_0, values = (lower_45_cast_fp16, var_2774_cast_fp16))[name = string("bottom_45_cast_fp16")]; + bool inv_59_interleave_0 = const()[name = string("inv_59_interleave_0"), val = bool(false)]; + tensor inv_59_cast_fp16 = concat(axis = var_2468, interleave = inv_59_interleave_0, values = (top_45_cast_fp16, bottom_45_cast_fp16))[name = string("inv_59_cast_fp16")]; + tensor var_2787 = const()[name = string("op_2787"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_2788_cast_fp16 = reshape(shape = var_2787, x = t_15_cast_fp16)[name = string("op_2788_cast_fp16")]; + tensor blocks_47_axes_0 = const()[name = string("blocks_47_axes_0"), val = tensor([-2])]; + bool blocks_47_keep_dims_0 = const()[name = string("blocks_47_keep_dims_0"), val = bool(false)]; + tensor blocks_47_cast_fp16 = reduce_sum(axes = blocks_47_axes_0, keep_dims = blocks_47_keep_dims_0, x = var_2788_cast_fp16)[name = string("blocks_47_cast_fp16")]; + tensor var_2796_begin_0 = const()[name = string("op_2796_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_2796_end_0 = const()[name = string("op_2796_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_2796_end_mask_0 = const()[name = string("op_2796_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2796_cast_fp16 = slice_by_index(begin = var_2796_begin_0, end = var_2796_end_0, end_mask = var_2796_end_mask_0, x = blocks_47_cast_fp16)[name = string("op_2796_cast_fp16")]; + tensor x_115_begin_0 = const()[name = string("x_115_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_115_end_0 = const()[name = string("x_115_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_115_end_mask_0 = const()[name = string("x_115_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_115_cast_fp16 = slice_by_index(begin = x_115_begin_0, end = x_115_end_0, end_mask = x_115_end_mask_0, x = var_2796_cast_fp16)[name = string("x_115_cast_fp16")]; + tensor var_2798 = const()[name = string("op_2798"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_55_cast_fp16 = reshape(shape = var_2798, x = inv_59_cast_fp16)[name = string("pair_55_cast_fp16")]; + tensor var_2800_begin_0 = const()[name = string("op_2800_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_2800_end_0 = const()[name = string("op_2800_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_2800_end_mask_0 = const()[name = string("op_2800_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2800_squeeze_mask_0 = const()[name = string("op_2800_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2800_cast_fp16 = slice_by_index(begin = var_2800_begin_0, end = var_2800_end_0, end_mask = var_2800_end_mask_0, squeeze_mask = var_2800_squeeze_mask_0, x = pair_55_cast_fp16)[name = string("op_2800_cast_fp16")]; + tensor var_2803_begin_0 = const()[name = string("op_2803_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_2803_end_0 = const()[name = string("op_2803_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_2803_end_mask_0 = const()[name = string("op_2803_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_2803_squeeze_mask_0 = const()[name = string("op_2803_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_2803_cast_fp16 = slice_by_index(begin = var_2803_begin_0, end = var_2803_end_0, end_mask = var_2803_end_mask_0, squeeze_mask = var_2803_squeeze_mask_0, x = pair_55_cast_fp16)[name = string("op_2803_cast_fp16")]; + bool var_2806_transpose_x_0 = const()[name = string("op_2806_transpose_x_0"), val = bool(false)]; + bool var_2806_transpose_y_0 = const()[name = string("op_2806_transpose_y_0"), val = bool(false)]; + tensor var_2806_cast_fp16 = matmul(transpose_x = var_2806_transpose_x_0, transpose_y = var_2806_transpose_y_0, x = x_115_cast_fp16, y = var_2800_cast_fp16)[name = string("op_2806_cast_fp16")]; + bool var_2807_transpose_x_0 = const()[name = string("op_2807_transpose_x_0"), val = bool(false)]; + bool var_2807_transpose_y_0 = const()[name = string("op_2807_transpose_y_0"), val = bool(false)]; + tensor var_2807_cast_fp16 = matmul(transpose_x = var_2807_transpose_x_0, transpose_y = var_2807_transpose_y_0, x = var_2803_cast_fp16, y = var_2806_cast_fp16)[name = string("op_2807_cast_fp16")]; + fp16 const_52_promoted_to_fp16 = const()[name = string("const_52_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_47_cast_fp16 = mul(x = var_2807_cast_fp16, y = const_52_promoted_to_fp16)[name = string("lower_47_cast_fp16")]; + bool top_47_interleave_0 = const()[name = string("top_47_interleave_0"), val = bool(false)]; + tensor top_47_cast_fp16 = concat(axis = var_2489, interleave = top_47_interleave_0, values = (var_2800_cast_fp16, var_378_to_fp16))[name = string("top_47_cast_fp16")]; + bool bottom_47_interleave_0 = const()[name = string("bottom_47_interleave_0"), val = bool(false)]; + tensor bottom_47_cast_fp16 = concat(axis = var_2489, interleave = bottom_47_interleave_0, values = (lower_47_cast_fp16, var_2803_cast_fp16))[name = string("bottom_47_cast_fp16")]; + bool inv_61_interleave_0 = const()[name = string("inv_61_interleave_0"), val = bool(false)]; + tensor inv_61_cast_fp16 = concat(axis = var_2468, interleave = inv_61_interleave_0, values = (top_47_cast_fp16, bottom_47_cast_fp16))[name = string("inv_61_cast_fp16")]; + tensor var_2816 = const()[name = string("op_2816"), val = tensor([32, 16, 64, 64])]; + tensor inv_63_cast_fp16 = reshape(shape = var_2816, x = inv_61_cast_fp16)[name = string("inv_63_cast_fp16")]; + bool new_v_7_transpose_x_0 = const()[name = string("new_v_7_transpose_x_0"), val = bool(false)]; + bool new_v_7_transpose_y_0 = const()[name = string("new_v_7_transpose_y_0"), val = bool(false)]; + tensor new_v_7_cast_fp16 = matmul(transpose_x = new_v_7_transpose_x_0, transpose_y = new_v_7_transpose_y_0, x = inv_63_cast_fp16, y = v_beta_7_cast_fp16)[name = string("new_v_7_cast_fp16")]; + tensor var_2819_cast_fp16 = exp(x = cum_7_cast_fp16)[name = string("op_2819_cast_fp16")]; + tensor var_2820_axes_0 = const()[name = string("op_2820_axes_0"), val = tensor([3])]; + tensor var_2820_cast_fp16 = expand_dims(axes = var_2820_axes_0, x = var_2819_cast_fp16)[name = string("op_2820_cast_fp16")]; + tensor var_2821_cast_fp16 = mul(x = k_beta_7_cast_fp16, y = var_2820_cast_fp16)[name = string("op_2821_cast_fp16")]; + bool k_cumdecay_7_transpose_x_0 = const()[name = string("k_cumdecay_7_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_7_transpose_y_0 = const()[name = string("k_cumdecay_7_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_7_cast_fp16 = matmul(transpose_x = k_cumdecay_7_transpose_x_0, transpose_y = k_cumdecay_7_transpose_y_0, x = inv_63_cast_fp16, y = var_2821_cast_fp16)[name = string("k_cumdecay_7_cast_fp16")]; + tensor q_dec_7_cast_fp16 = mul(x = q_43_cast_fp16, y = var_2820_cast_fp16)[name = string("q_dec_7_cast_fp16")]; + tensor var_2826_cast_fp16 = exp(x = to_end_7_cast_fp16)[name = string("op_2826_cast_fp16")]; + tensor var_2827_axes_0 = const()[name = string("op_2827_axes_0"), val = tensor([3])]; + tensor var_2827_cast_fp16 = expand_dims(axes = var_2827_axes_0, x = var_2826_cast_fp16)[name = string("op_2827_cast_fp16")]; + tensor k_dec_7_cast_fp16 = mul(x = k_37_cast_fp16, y = var_2827_cast_fp16)[name = string("k_dec_7_cast_fp16")]; + tensor var_2829_begin_0 = const()[name = string("op_2829_begin_0"), val = tensor([0, 0, -1])]; + tensor var_2829_end_0 = const()[name = string("op_2829_end_0"), val = tensor([32, 16, 64])]; + tensor var_2829_end_mask_0 = const()[name = string("op_2829_end_mask_0"), val = tensor([true, true, true])]; + tensor var_2829_cast_fp16 = slice_by_index(begin = var_2829_begin_0, end = var_2829_end_0, end_mask = var_2829_end_mask_0, x = cum_7_cast_fp16)[name = string("op_2829_cast_fp16")]; + tensor var_2830_cast_fp16 = exp(x = var_2829_cast_fp16)[name = string("op_2830_cast_fp16")]; + tensor chunk_decay_7_axes_0 = const()[name = string("chunk_decay_7_axes_0"), val = tensor([3])]; + tensor chunk_decay_7_cast_fp16 = expand_dims(axes = chunk_decay_7_axes_0, x = var_2830_cast_fp16)[name = string("chunk_decay_7_cast_fp16")]; + tensor var_2835_begin_0 = const()[name = string("op_2835_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2835_end_0 = const()[name = string("op_2835_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_2835_end_mask_0 = const()[name = string("op_2835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2835_squeeze_mask_0 = const()[name = string("op_2835_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2835_cast_fp16 = slice_by_index(begin = var_2835_begin_0, end = var_2835_end_0, end_mask = var_2835_end_mask_0, squeeze_mask = var_2835_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2835_cast_fp16")]; + tensor var_2837_begin_0 = const()[name = string("op_2837_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2837_end_0 = const()[name = string("op_2837_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_2837_end_mask_0 = const()[name = string("op_2837_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2837_squeeze_mask_0 = const()[name = string("op_2837_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2837_cast_fp16 = slice_by_index(begin = var_2837_begin_0, end = var_2837_end_0, end_mask = var_2837_end_mask_0, squeeze_mask = var_2837_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2837_cast_fp16")]; + bool var_2838_transpose_x_0 = const()[name = string("op_2838_transpose_x_0"), val = bool(false)]; + bool var_2838_transpose_y_0 = const()[name = string("op_2838_transpose_y_0"), val = bool(false)]; + tensor var_2838_cast_fp16 = matmul(transpose_x = var_2838_transpose_x_0, transpose_y = var_2838_transpose_y_0, x = var_2837_cast_fp16, y = state_1_to_fp16)[name = string("op_2838_cast_fp16")]; + tensor v_new_97_cast_fp16 = sub(x = var_2835_cast_fp16, y = var_2838_cast_fp16)[name = string("v_new_97_cast_fp16")]; + tensor var_2841_begin_0 = const()[name = string("op_2841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2841_end_0 = const()[name = string("op_2841_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_2841_end_mask_0 = const()[name = string("op_2841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2841_squeeze_mask_0 = const()[name = string("op_2841_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2841_cast_fp16 = slice_by_index(begin = var_2841_begin_0, end = var_2841_end_0, end_mask = var_2841_end_mask_0, squeeze_mask = var_2841_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2841_cast_fp16")]; + bool var_2842_transpose_x_0 = const()[name = string("op_2842_transpose_x_0"), val = bool(false)]; + bool var_2842_transpose_y_0 = const()[name = string("op_2842_transpose_y_0"), val = bool(false)]; + tensor var_2842_cast_fp16 = matmul(transpose_x = var_2842_transpose_x_0, transpose_y = var_2842_transpose_y_0, x = var_2841_cast_fp16, y = state_1_to_fp16)[name = string("op_2842_cast_fp16")]; + tensor var_2844_begin_0 = const()[name = string("op_2844_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2844_end_0 = const()[name = string("op_2844_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_2844_end_mask_0 = const()[name = string("op_2844_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2844_squeeze_mask_0 = const()[name = string("op_2844_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2844_cast_fp16 = slice_by_index(begin = var_2844_begin_0, end = var_2844_end_0, end_mask = var_2844_end_mask_0, squeeze_mask = var_2844_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2844_cast_fp16")]; + bool var_2845_transpose_x_0 = const()[name = string("op_2845_transpose_x_0"), val = bool(false)]; + bool var_2845_transpose_y_0 = const()[name = string("op_2845_transpose_y_0"), val = bool(false)]; + tensor var_2845_cast_fp16 = matmul(transpose_x = var_2845_transpose_x_0, transpose_y = var_2845_transpose_y_0, x = var_2844_cast_fp16, y = v_new_97_cast_fp16)[name = string("op_2845_cast_fp16")]; + tensor var_2846_cast_fp16 = add(x = var_2842_cast_fp16, y = var_2845_cast_fp16)[name = string("op_2846_cast_fp16")]; + tensor var_2848_begin_0 = const()[name = string("op_2848_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2848_end_0 = const()[name = string("op_2848_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_2848_end_mask_0 = const()[name = string("op_2848_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2848_squeeze_mask_0 = const()[name = string("op_2848_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2848_cast_fp16 = slice_by_index(begin = var_2848_begin_0, end = var_2848_end_0, end_mask = var_2848_end_mask_0, squeeze_mask = var_2848_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2848_cast_fp16")]; + tensor var_2849_cast_fp16 = mul(x = state_1_to_fp16, y = var_2848_cast_fp16)[name = string("op_2849_cast_fp16")]; + tensor var_2851_begin_0 = const()[name = string("op_2851_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2851_end_0 = const()[name = string("op_2851_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_2851_end_mask_0 = const()[name = string("op_2851_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2851_squeeze_mask_0 = const()[name = string("op_2851_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2851_cast_fp16 = slice_by_index(begin = var_2851_begin_0, end = var_2851_end_0, end_mask = var_2851_end_mask_0, squeeze_mask = var_2851_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2851_cast_fp16")]; + bool var_2853_transpose_x_1 = const()[name = string("op_2853_transpose_x_1"), val = bool(true)]; + bool var_2853_transpose_y_1 = const()[name = string("op_2853_transpose_y_1"), val = bool(false)]; + tensor var_2853_cast_fp16 = matmul(transpose_x = var_2853_transpose_x_1, transpose_y = var_2853_transpose_y_1, x = var_2851_cast_fp16, y = v_new_97_cast_fp16)[name = string("op_2853_cast_fp16")]; + tensor state_99_cast_fp16 = add(x = var_2849_cast_fp16, y = var_2853_cast_fp16)[name = string("state_99_cast_fp16")]; + tensor var_2856_begin_0 = const()[name = string("op_2856_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2856_end_0 = const()[name = string("op_2856_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_2856_end_mask_0 = const()[name = string("op_2856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2856_squeeze_mask_0 = const()[name = string("op_2856_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2856_cast_fp16 = slice_by_index(begin = var_2856_begin_0, end = var_2856_end_0, end_mask = var_2856_end_mask_0, squeeze_mask = var_2856_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2856_cast_fp16")]; + tensor var_2858_begin_0 = const()[name = string("op_2858_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2858_end_0 = const()[name = string("op_2858_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_2858_end_mask_0 = const()[name = string("op_2858_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2858_squeeze_mask_0 = const()[name = string("op_2858_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2858_cast_fp16 = slice_by_index(begin = var_2858_begin_0, end = var_2858_end_0, end_mask = var_2858_end_mask_0, squeeze_mask = var_2858_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2858_cast_fp16")]; + bool var_2859_transpose_x_0 = const()[name = string("op_2859_transpose_x_0"), val = bool(false)]; + bool var_2859_transpose_y_0 = const()[name = string("op_2859_transpose_y_0"), val = bool(false)]; + tensor var_2859_cast_fp16 = matmul(transpose_x = var_2859_transpose_x_0, transpose_y = var_2859_transpose_y_0, x = var_2858_cast_fp16, y = state_99_cast_fp16)[name = string("op_2859_cast_fp16")]; + tensor v_new_99_cast_fp16 = sub(x = var_2856_cast_fp16, y = var_2859_cast_fp16)[name = string("v_new_99_cast_fp16")]; + tensor var_2862_begin_0 = const()[name = string("op_2862_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2862_end_0 = const()[name = string("op_2862_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_2862_end_mask_0 = const()[name = string("op_2862_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2862_squeeze_mask_0 = const()[name = string("op_2862_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2862_cast_fp16 = slice_by_index(begin = var_2862_begin_0, end = var_2862_end_0, end_mask = var_2862_end_mask_0, squeeze_mask = var_2862_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2862_cast_fp16")]; + bool var_2863_transpose_x_0 = const()[name = string("op_2863_transpose_x_0"), val = bool(false)]; + bool var_2863_transpose_y_0 = const()[name = string("op_2863_transpose_y_0"), val = bool(false)]; + tensor var_2863_cast_fp16 = matmul(transpose_x = var_2863_transpose_x_0, transpose_y = var_2863_transpose_y_0, x = var_2862_cast_fp16, y = state_99_cast_fp16)[name = string("op_2863_cast_fp16")]; + tensor var_2865_begin_0 = const()[name = string("op_2865_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2865_end_0 = const()[name = string("op_2865_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_2865_end_mask_0 = const()[name = string("op_2865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2865_squeeze_mask_0 = const()[name = string("op_2865_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2865_cast_fp16 = slice_by_index(begin = var_2865_begin_0, end = var_2865_end_0, end_mask = var_2865_end_mask_0, squeeze_mask = var_2865_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2865_cast_fp16")]; + bool var_2866_transpose_x_0 = const()[name = string("op_2866_transpose_x_0"), val = bool(false)]; + bool var_2866_transpose_y_0 = const()[name = string("op_2866_transpose_y_0"), val = bool(false)]; + tensor var_2866_cast_fp16 = matmul(transpose_x = var_2866_transpose_x_0, transpose_y = var_2866_transpose_y_0, x = var_2865_cast_fp16, y = v_new_99_cast_fp16)[name = string("op_2866_cast_fp16")]; + tensor var_2867_cast_fp16 = add(x = var_2863_cast_fp16, y = var_2866_cast_fp16)[name = string("op_2867_cast_fp16")]; + tensor var_2869_begin_0 = const()[name = string("op_2869_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2869_end_0 = const()[name = string("op_2869_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_2869_end_mask_0 = const()[name = string("op_2869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2869_squeeze_mask_0 = const()[name = string("op_2869_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2869_cast_fp16 = slice_by_index(begin = var_2869_begin_0, end = var_2869_end_0, end_mask = var_2869_end_mask_0, squeeze_mask = var_2869_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2869_cast_fp16")]; + tensor var_2870_cast_fp16 = mul(x = state_99_cast_fp16, y = var_2869_cast_fp16)[name = string("op_2870_cast_fp16")]; + tensor var_2872_begin_0 = const()[name = string("op_2872_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_2872_end_0 = const()[name = string("op_2872_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_2872_end_mask_0 = const()[name = string("op_2872_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2872_squeeze_mask_0 = const()[name = string("op_2872_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2872_cast_fp16 = slice_by_index(begin = var_2872_begin_0, end = var_2872_end_0, end_mask = var_2872_end_mask_0, squeeze_mask = var_2872_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2872_cast_fp16")]; + bool var_2874_transpose_x_1 = const()[name = string("op_2874_transpose_x_1"), val = bool(true)]; + bool var_2874_transpose_y_1 = const()[name = string("op_2874_transpose_y_1"), val = bool(false)]; + tensor var_2874_cast_fp16 = matmul(transpose_x = var_2874_transpose_x_1, transpose_y = var_2874_transpose_y_1, x = var_2872_cast_fp16, y = v_new_99_cast_fp16)[name = string("op_2874_cast_fp16")]; + tensor state_101_cast_fp16 = add(x = var_2870_cast_fp16, y = var_2874_cast_fp16)[name = string("state_101_cast_fp16")]; + tensor var_2877_begin_0 = const()[name = string("op_2877_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2877_end_0 = const()[name = string("op_2877_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_2877_end_mask_0 = const()[name = string("op_2877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2877_squeeze_mask_0 = const()[name = string("op_2877_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2877_cast_fp16 = slice_by_index(begin = var_2877_begin_0, end = var_2877_end_0, end_mask = var_2877_end_mask_0, squeeze_mask = var_2877_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2877_cast_fp16")]; + tensor var_2879_begin_0 = const()[name = string("op_2879_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2879_end_0 = const()[name = string("op_2879_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_2879_end_mask_0 = const()[name = string("op_2879_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2879_squeeze_mask_0 = const()[name = string("op_2879_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2879_cast_fp16 = slice_by_index(begin = var_2879_begin_0, end = var_2879_end_0, end_mask = var_2879_end_mask_0, squeeze_mask = var_2879_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2879_cast_fp16")]; + bool var_2880_transpose_x_0 = const()[name = string("op_2880_transpose_x_0"), val = bool(false)]; + bool var_2880_transpose_y_0 = const()[name = string("op_2880_transpose_y_0"), val = bool(false)]; + tensor var_2880_cast_fp16 = matmul(transpose_x = var_2880_transpose_x_0, transpose_y = var_2880_transpose_y_0, x = var_2879_cast_fp16, y = state_101_cast_fp16)[name = string("op_2880_cast_fp16")]; + tensor v_new_101_cast_fp16 = sub(x = var_2877_cast_fp16, y = var_2880_cast_fp16)[name = string("v_new_101_cast_fp16")]; + tensor var_2883_begin_0 = const()[name = string("op_2883_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2883_end_0 = const()[name = string("op_2883_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_2883_end_mask_0 = const()[name = string("op_2883_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2883_squeeze_mask_0 = const()[name = string("op_2883_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2883_cast_fp16 = slice_by_index(begin = var_2883_begin_0, end = var_2883_end_0, end_mask = var_2883_end_mask_0, squeeze_mask = var_2883_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2883_cast_fp16")]; + bool var_2884_transpose_x_0 = const()[name = string("op_2884_transpose_x_0"), val = bool(false)]; + bool var_2884_transpose_y_0 = const()[name = string("op_2884_transpose_y_0"), val = bool(false)]; + tensor var_2884_cast_fp16 = matmul(transpose_x = var_2884_transpose_x_0, transpose_y = var_2884_transpose_y_0, x = var_2883_cast_fp16, y = state_101_cast_fp16)[name = string("op_2884_cast_fp16")]; + tensor var_2886_begin_0 = const()[name = string("op_2886_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2886_end_0 = const()[name = string("op_2886_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_2886_end_mask_0 = const()[name = string("op_2886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2886_squeeze_mask_0 = const()[name = string("op_2886_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2886_cast_fp16 = slice_by_index(begin = var_2886_begin_0, end = var_2886_end_0, end_mask = var_2886_end_mask_0, squeeze_mask = var_2886_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2886_cast_fp16")]; + bool var_2887_transpose_x_0 = const()[name = string("op_2887_transpose_x_0"), val = bool(false)]; + bool var_2887_transpose_y_0 = const()[name = string("op_2887_transpose_y_0"), val = bool(false)]; + tensor var_2887_cast_fp16 = matmul(transpose_x = var_2887_transpose_x_0, transpose_y = var_2887_transpose_y_0, x = var_2886_cast_fp16, y = v_new_101_cast_fp16)[name = string("op_2887_cast_fp16")]; + tensor var_2888_cast_fp16 = add(x = var_2884_cast_fp16, y = var_2887_cast_fp16)[name = string("op_2888_cast_fp16")]; + tensor var_2890_begin_0 = const()[name = string("op_2890_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2890_end_0 = const()[name = string("op_2890_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_2890_end_mask_0 = const()[name = string("op_2890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2890_squeeze_mask_0 = const()[name = string("op_2890_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2890_cast_fp16 = slice_by_index(begin = var_2890_begin_0, end = var_2890_end_0, end_mask = var_2890_end_mask_0, squeeze_mask = var_2890_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2890_cast_fp16")]; + tensor var_2891_cast_fp16 = mul(x = state_101_cast_fp16, y = var_2890_cast_fp16)[name = string("op_2891_cast_fp16")]; + tensor var_2893_begin_0 = const()[name = string("op_2893_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_2893_end_0 = const()[name = string("op_2893_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_2893_end_mask_0 = const()[name = string("op_2893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2893_squeeze_mask_0 = const()[name = string("op_2893_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2893_cast_fp16 = slice_by_index(begin = var_2893_begin_0, end = var_2893_end_0, end_mask = var_2893_end_mask_0, squeeze_mask = var_2893_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2893_cast_fp16")]; + bool var_2895_transpose_x_1 = const()[name = string("op_2895_transpose_x_1"), val = bool(true)]; + bool var_2895_transpose_y_1 = const()[name = string("op_2895_transpose_y_1"), val = bool(false)]; + tensor var_2895_cast_fp16 = matmul(transpose_x = var_2895_transpose_x_1, transpose_y = var_2895_transpose_y_1, x = var_2893_cast_fp16, y = v_new_101_cast_fp16)[name = string("op_2895_cast_fp16")]; + tensor state_103_cast_fp16 = add(x = var_2891_cast_fp16, y = var_2895_cast_fp16)[name = string("state_103_cast_fp16")]; + tensor var_2898_begin_0 = const()[name = string("op_2898_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2898_end_0 = const()[name = string("op_2898_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_2898_end_mask_0 = const()[name = string("op_2898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2898_squeeze_mask_0 = const()[name = string("op_2898_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2898_cast_fp16 = slice_by_index(begin = var_2898_begin_0, end = var_2898_end_0, end_mask = var_2898_end_mask_0, squeeze_mask = var_2898_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2898_cast_fp16")]; + tensor var_2900_begin_0 = const()[name = string("op_2900_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2900_end_0 = const()[name = string("op_2900_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_2900_end_mask_0 = const()[name = string("op_2900_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2900_squeeze_mask_0 = const()[name = string("op_2900_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2900_cast_fp16 = slice_by_index(begin = var_2900_begin_0, end = var_2900_end_0, end_mask = var_2900_end_mask_0, squeeze_mask = var_2900_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2900_cast_fp16")]; + bool var_2901_transpose_x_0 = const()[name = string("op_2901_transpose_x_0"), val = bool(false)]; + bool var_2901_transpose_y_0 = const()[name = string("op_2901_transpose_y_0"), val = bool(false)]; + tensor var_2901_cast_fp16 = matmul(transpose_x = var_2901_transpose_x_0, transpose_y = var_2901_transpose_y_0, x = var_2900_cast_fp16, y = state_103_cast_fp16)[name = string("op_2901_cast_fp16")]; + tensor v_new_103_cast_fp16 = sub(x = var_2898_cast_fp16, y = var_2901_cast_fp16)[name = string("v_new_103_cast_fp16")]; + tensor var_2904_begin_0 = const()[name = string("op_2904_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2904_end_0 = const()[name = string("op_2904_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_2904_end_mask_0 = const()[name = string("op_2904_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2904_squeeze_mask_0 = const()[name = string("op_2904_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2904_cast_fp16 = slice_by_index(begin = var_2904_begin_0, end = var_2904_end_0, end_mask = var_2904_end_mask_0, squeeze_mask = var_2904_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2904_cast_fp16")]; + bool var_2905_transpose_x_0 = const()[name = string("op_2905_transpose_x_0"), val = bool(false)]; + bool var_2905_transpose_y_0 = const()[name = string("op_2905_transpose_y_0"), val = bool(false)]; + tensor var_2905_cast_fp16 = matmul(transpose_x = var_2905_transpose_x_0, transpose_y = var_2905_transpose_y_0, x = var_2904_cast_fp16, y = state_103_cast_fp16)[name = string("op_2905_cast_fp16")]; + tensor var_2907_begin_0 = const()[name = string("op_2907_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2907_end_0 = const()[name = string("op_2907_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_2907_end_mask_0 = const()[name = string("op_2907_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2907_squeeze_mask_0 = const()[name = string("op_2907_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2907_cast_fp16 = slice_by_index(begin = var_2907_begin_0, end = var_2907_end_0, end_mask = var_2907_end_mask_0, squeeze_mask = var_2907_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2907_cast_fp16")]; + bool var_2908_transpose_x_0 = const()[name = string("op_2908_transpose_x_0"), val = bool(false)]; + bool var_2908_transpose_y_0 = const()[name = string("op_2908_transpose_y_0"), val = bool(false)]; + tensor var_2908_cast_fp16 = matmul(transpose_x = var_2908_transpose_x_0, transpose_y = var_2908_transpose_y_0, x = var_2907_cast_fp16, y = v_new_103_cast_fp16)[name = string("op_2908_cast_fp16")]; + tensor var_2909_cast_fp16 = add(x = var_2905_cast_fp16, y = var_2908_cast_fp16)[name = string("op_2909_cast_fp16")]; + tensor var_2911_begin_0 = const()[name = string("op_2911_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2911_end_0 = const()[name = string("op_2911_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_2911_end_mask_0 = const()[name = string("op_2911_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2911_squeeze_mask_0 = const()[name = string("op_2911_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2911_cast_fp16 = slice_by_index(begin = var_2911_begin_0, end = var_2911_end_0, end_mask = var_2911_end_mask_0, squeeze_mask = var_2911_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2911_cast_fp16")]; + tensor var_2912_cast_fp16 = mul(x = state_103_cast_fp16, y = var_2911_cast_fp16)[name = string("op_2912_cast_fp16")]; + tensor var_2914_begin_0 = const()[name = string("op_2914_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_2914_end_0 = const()[name = string("op_2914_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_2914_end_mask_0 = const()[name = string("op_2914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2914_squeeze_mask_0 = const()[name = string("op_2914_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2914_cast_fp16 = slice_by_index(begin = var_2914_begin_0, end = var_2914_end_0, end_mask = var_2914_end_mask_0, squeeze_mask = var_2914_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2914_cast_fp16")]; + bool var_2916_transpose_x_1 = const()[name = string("op_2916_transpose_x_1"), val = bool(true)]; + bool var_2916_transpose_y_1 = const()[name = string("op_2916_transpose_y_1"), val = bool(false)]; + tensor var_2916_cast_fp16 = matmul(transpose_x = var_2916_transpose_x_1, transpose_y = var_2916_transpose_y_1, x = var_2914_cast_fp16, y = v_new_103_cast_fp16)[name = string("op_2916_cast_fp16")]; + tensor state_105_cast_fp16 = add(x = var_2912_cast_fp16, y = var_2916_cast_fp16)[name = string("state_105_cast_fp16")]; + tensor var_2919_begin_0 = const()[name = string("op_2919_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2919_end_0 = const()[name = string("op_2919_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2919_end_mask_0 = const()[name = string("op_2919_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2919_squeeze_mask_0 = const()[name = string("op_2919_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2919_cast_fp16 = slice_by_index(begin = var_2919_begin_0, end = var_2919_end_0, end_mask = var_2919_end_mask_0, squeeze_mask = var_2919_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2919_cast_fp16")]; + tensor var_2921_begin_0 = const()[name = string("op_2921_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2921_end_0 = const()[name = string("op_2921_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2921_end_mask_0 = const()[name = string("op_2921_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2921_squeeze_mask_0 = const()[name = string("op_2921_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2921_cast_fp16 = slice_by_index(begin = var_2921_begin_0, end = var_2921_end_0, end_mask = var_2921_end_mask_0, squeeze_mask = var_2921_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2921_cast_fp16")]; + bool var_2922_transpose_x_0 = const()[name = string("op_2922_transpose_x_0"), val = bool(false)]; + bool var_2922_transpose_y_0 = const()[name = string("op_2922_transpose_y_0"), val = bool(false)]; + tensor var_2922_cast_fp16 = matmul(transpose_x = var_2922_transpose_x_0, transpose_y = var_2922_transpose_y_0, x = var_2921_cast_fp16, y = state_105_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor v_new_105_cast_fp16 = sub(x = var_2919_cast_fp16, y = var_2922_cast_fp16)[name = string("v_new_105_cast_fp16")]; + tensor var_2925_begin_0 = const()[name = string("op_2925_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2925_end_0 = const()[name = string("op_2925_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2925_end_mask_0 = const()[name = string("op_2925_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2925_squeeze_mask_0 = const()[name = string("op_2925_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2925_cast_fp16 = slice_by_index(begin = var_2925_begin_0, end = var_2925_end_0, end_mask = var_2925_end_mask_0, squeeze_mask = var_2925_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2925_cast_fp16")]; + bool var_2926_transpose_x_0 = const()[name = string("op_2926_transpose_x_0"), val = bool(false)]; + bool var_2926_transpose_y_0 = const()[name = string("op_2926_transpose_y_0"), val = bool(false)]; + tensor var_2926_cast_fp16 = matmul(transpose_x = var_2926_transpose_x_0, transpose_y = var_2926_transpose_y_0, x = var_2925_cast_fp16, y = state_105_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2928_begin_0 = const()[name = string("op_2928_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2928_end_0 = const()[name = string("op_2928_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_2928_end_mask_0 = const()[name = string("op_2928_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2928_squeeze_mask_0 = const()[name = string("op_2928_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2928_cast_fp16 = slice_by_index(begin = var_2928_begin_0, end = var_2928_end_0, end_mask = var_2928_end_mask_0, squeeze_mask = var_2928_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2928_cast_fp16")]; + bool var_2929_transpose_x_0 = const()[name = string("op_2929_transpose_x_0"), val = bool(false)]; + bool var_2929_transpose_y_0 = const()[name = string("op_2929_transpose_y_0"), val = bool(false)]; + tensor var_2929_cast_fp16 = matmul(transpose_x = var_2929_transpose_x_0, transpose_y = var_2929_transpose_y_0, x = var_2928_cast_fp16, y = v_new_105_cast_fp16)[name = string("op_2929_cast_fp16")]; + tensor var_2930_cast_fp16 = add(x = var_2926_cast_fp16, y = var_2929_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2932_begin_0 = const()[name = string("op_2932_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2932_end_0 = const()[name = string("op_2932_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_2932_end_mask_0 = const()[name = string("op_2932_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2932_squeeze_mask_0 = const()[name = string("op_2932_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2932_cast_fp16 = slice_by_index(begin = var_2932_begin_0, end = var_2932_end_0, end_mask = var_2932_end_mask_0, squeeze_mask = var_2932_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2932_cast_fp16")]; + tensor var_2933_cast_fp16 = mul(x = state_105_cast_fp16, y = var_2932_cast_fp16)[name = string("op_2933_cast_fp16")]; + tensor var_2935_begin_0 = const()[name = string("op_2935_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_2935_end_0 = const()[name = string("op_2935_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_2935_end_mask_0 = const()[name = string("op_2935_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2935_squeeze_mask_0 = const()[name = string("op_2935_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2935_cast_fp16 = slice_by_index(begin = var_2935_begin_0, end = var_2935_end_0, end_mask = var_2935_end_mask_0, squeeze_mask = var_2935_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2935_cast_fp16")]; + bool var_2937_transpose_x_1 = const()[name = string("op_2937_transpose_x_1"), val = bool(true)]; + bool var_2937_transpose_y_1 = const()[name = string("op_2937_transpose_y_1"), val = bool(false)]; + tensor var_2937_cast_fp16 = matmul(transpose_x = var_2937_transpose_x_1, transpose_y = var_2937_transpose_y_1, x = var_2935_cast_fp16, y = v_new_105_cast_fp16)[name = string("op_2937_cast_fp16")]; + tensor state_107_cast_fp16 = add(x = var_2933_cast_fp16, y = var_2937_cast_fp16)[name = string("state_107_cast_fp16")]; + tensor var_2940_begin_0 = const()[name = string("op_2940_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2940_end_0 = const()[name = string("op_2940_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2940_end_mask_0 = const()[name = string("op_2940_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2940_squeeze_mask_0 = const()[name = string("op_2940_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2940_cast_fp16 = slice_by_index(begin = var_2940_begin_0, end = var_2940_end_0, end_mask = var_2940_end_mask_0, squeeze_mask = var_2940_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2940_cast_fp16")]; + tensor var_2942_begin_0 = const()[name = string("op_2942_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2942_end_0 = const()[name = string("op_2942_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2942_end_mask_0 = const()[name = string("op_2942_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2942_squeeze_mask_0 = const()[name = string("op_2942_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2942_cast_fp16 = slice_by_index(begin = var_2942_begin_0, end = var_2942_end_0, end_mask = var_2942_end_mask_0, squeeze_mask = var_2942_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2942_cast_fp16")]; + bool var_2943_transpose_x_0 = const()[name = string("op_2943_transpose_x_0"), val = bool(false)]; + bool var_2943_transpose_y_0 = const()[name = string("op_2943_transpose_y_0"), val = bool(false)]; + tensor var_2943_cast_fp16 = matmul(transpose_x = var_2943_transpose_x_0, transpose_y = var_2943_transpose_y_0, x = var_2942_cast_fp16, y = state_107_cast_fp16)[name = string("op_2943_cast_fp16")]; + tensor v_new_107_cast_fp16 = sub(x = var_2940_cast_fp16, y = var_2943_cast_fp16)[name = string("v_new_107_cast_fp16")]; + tensor var_2946_begin_0 = const()[name = string("op_2946_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2946_end_0 = const()[name = string("op_2946_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2946_end_mask_0 = const()[name = string("op_2946_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2946_squeeze_mask_0 = const()[name = string("op_2946_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2946_cast_fp16 = slice_by_index(begin = var_2946_begin_0, end = var_2946_end_0, end_mask = var_2946_end_mask_0, squeeze_mask = var_2946_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2946_cast_fp16")]; + bool var_2947_transpose_x_0 = const()[name = string("op_2947_transpose_x_0"), val = bool(false)]; + bool var_2947_transpose_y_0 = const()[name = string("op_2947_transpose_y_0"), val = bool(false)]; + tensor var_2947_cast_fp16 = matmul(transpose_x = var_2947_transpose_x_0, transpose_y = var_2947_transpose_y_0, x = var_2946_cast_fp16, y = state_107_cast_fp16)[name = string("op_2947_cast_fp16")]; + tensor var_2949_begin_0 = const()[name = string("op_2949_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2949_end_0 = const()[name = string("op_2949_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_2949_end_mask_0 = const()[name = string("op_2949_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2949_squeeze_mask_0 = const()[name = string("op_2949_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2949_cast_fp16 = slice_by_index(begin = var_2949_begin_0, end = var_2949_end_0, end_mask = var_2949_end_mask_0, squeeze_mask = var_2949_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2949_cast_fp16")]; + bool var_2950_transpose_x_0 = const()[name = string("op_2950_transpose_x_0"), val = bool(false)]; + bool var_2950_transpose_y_0 = const()[name = string("op_2950_transpose_y_0"), val = bool(false)]; + tensor var_2950_cast_fp16 = matmul(transpose_x = var_2950_transpose_x_0, transpose_y = var_2950_transpose_y_0, x = var_2949_cast_fp16, y = v_new_107_cast_fp16)[name = string("op_2950_cast_fp16")]; + tensor var_2951_cast_fp16 = add(x = var_2947_cast_fp16, y = var_2950_cast_fp16)[name = string("op_2951_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = string("op_2953_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2953_end_0 = const()[name = string("op_2953_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_2953_end_mask_0 = const()[name = string("op_2953_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2953_squeeze_mask_0 = const()[name = string("op_2953_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, squeeze_mask = var_2953_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2954_cast_fp16 = mul(x = state_107_cast_fp16, y = var_2953_cast_fp16)[name = string("op_2954_cast_fp16")]; + tensor var_2956_begin_0 = const()[name = string("op_2956_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_2956_end_0 = const()[name = string("op_2956_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_2956_end_mask_0 = const()[name = string("op_2956_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2956_squeeze_mask_0 = const()[name = string("op_2956_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2956_cast_fp16 = slice_by_index(begin = var_2956_begin_0, end = var_2956_end_0, end_mask = var_2956_end_mask_0, squeeze_mask = var_2956_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2956_cast_fp16")]; + bool var_2958_transpose_x_1 = const()[name = string("op_2958_transpose_x_1"), val = bool(true)]; + bool var_2958_transpose_y_1 = const()[name = string("op_2958_transpose_y_1"), val = bool(false)]; + tensor var_2958_cast_fp16 = matmul(transpose_x = var_2958_transpose_x_1, transpose_y = var_2958_transpose_y_1, x = var_2956_cast_fp16, y = v_new_107_cast_fp16)[name = string("op_2958_cast_fp16")]; + tensor state_109_cast_fp16 = add(x = var_2954_cast_fp16, y = var_2958_cast_fp16)[name = string("state_109_cast_fp16")]; + tensor var_2961_begin_0 = const()[name = string("op_2961_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2961_end_0 = const()[name = string("op_2961_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2961_end_mask_0 = const()[name = string("op_2961_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2961_squeeze_mask_0 = const()[name = string("op_2961_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2961_cast_fp16 = slice_by_index(begin = var_2961_begin_0, end = var_2961_end_0, end_mask = var_2961_end_mask_0, squeeze_mask = var_2961_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2961_cast_fp16")]; + tensor var_2963_begin_0 = const()[name = string("op_2963_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2963_end_0 = const()[name = string("op_2963_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2963_end_mask_0 = const()[name = string("op_2963_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2963_squeeze_mask_0 = const()[name = string("op_2963_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2963_cast_fp16 = slice_by_index(begin = var_2963_begin_0, end = var_2963_end_0, end_mask = var_2963_end_mask_0, squeeze_mask = var_2963_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2963_cast_fp16")]; + bool var_2964_transpose_x_0 = const()[name = string("op_2964_transpose_x_0"), val = bool(false)]; + bool var_2964_transpose_y_0 = const()[name = string("op_2964_transpose_y_0"), val = bool(false)]; + tensor var_2964_cast_fp16 = matmul(transpose_x = var_2964_transpose_x_0, transpose_y = var_2964_transpose_y_0, x = var_2963_cast_fp16, y = state_109_cast_fp16)[name = string("op_2964_cast_fp16")]; + tensor v_new_109_cast_fp16 = sub(x = var_2961_cast_fp16, y = var_2964_cast_fp16)[name = string("v_new_109_cast_fp16")]; + tensor var_2967_begin_0 = const()[name = string("op_2967_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2967_end_0 = const()[name = string("op_2967_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2967_end_mask_0 = const()[name = string("op_2967_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2967_squeeze_mask_0 = const()[name = string("op_2967_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2967_cast_fp16 = slice_by_index(begin = var_2967_begin_0, end = var_2967_end_0, end_mask = var_2967_end_mask_0, squeeze_mask = var_2967_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2967_cast_fp16")]; + bool var_2968_transpose_x_0 = const()[name = string("op_2968_transpose_x_0"), val = bool(false)]; + bool var_2968_transpose_y_0 = const()[name = string("op_2968_transpose_y_0"), val = bool(false)]; + tensor var_2968_cast_fp16 = matmul(transpose_x = var_2968_transpose_x_0, transpose_y = var_2968_transpose_y_0, x = var_2967_cast_fp16, y = state_109_cast_fp16)[name = string("op_2968_cast_fp16")]; + tensor var_2970_begin_0 = const()[name = string("op_2970_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2970_end_0 = const()[name = string("op_2970_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_2970_end_mask_0 = const()[name = string("op_2970_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2970_squeeze_mask_0 = const()[name = string("op_2970_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2970_cast_fp16 = slice_by_index(begin = var_2970_begin_0, end = var_2970_end_0, end_mask = var_2970_end_mask_0, squeeze_mask = var_2970_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2970_cast_fp16")]; + bool var_2971_transpose_x_0 = const()[name = string("op_2971_transpose_x_0"), val = bool(false)]; + bool var_2971_transpose_y_0 = const()[name = string("op_2971_transpose_y_0"), val = bool(false)]; + tensor var_2971_cast_fp16 = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2970_cast_fp16, y = v_new_109_cast_fp16)[name = string("op_2971_cast_fp16")]; + tensor var_2972_cast_fp16 = add(x = var_2968_cast_fp16, y = var_2971_cast_fp16)[name = string("op_2972_cast_fp16")]; + tensor var_2974_begin_0 = const()[name = string("op_2974_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2974_end_0 = const()[name = string("op_2974_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_2974_end_mask_0 = const()[name = string("op_2974_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2974_squeeze_mask_0 = const()[name = string("op_2974_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2974_cast_fp16 = slice_by_index(begin = var_2974_begin_0, end = var_2974_end_0, end_mask = var_2974_end_mask_0, squeeze_mask = var_2974_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2975_cast_fp16 = mul(x = state_109_cast_fp16, y = var_2974_cast_fp16)[name = string("op_2975_cast_fp16")]; + tensor var_2977_begin_0 = const()[name = string("op_2977_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_2977_end_0 = const()[name = string("op_2977_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_2977_end_mask_0 = const()[name = string("op_2977_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2977_squeeze_mask_0 = const()[name = string("op_2977_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2977_cast_fp16 = slice_by_index(begin = var_2977_begin_0, end = var_2977_end_0, end_mask = var_2977_end_mask_0, squeeze_mask = var_2977_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2977_cast_fp16")]; + bool var_2979_transpose_x_1 = const()[name = string("op_2979_transpose_x_1"), val = bool(true)]; + bool var_2979_transpose_y_1 = const()[name = string("op_2979_transpose_y_1"), val = bool(false)]; + tensor var_2979_cast_fp16 = matmul(transpose_x = var_2979_transpose_x_1, transpose_y = var_2979_transpose_y_1, x = var_2977_cast_fp16, y = v_new_109_cast_fp16)[name = string("op_2979_cast_fp16")]; + tensor state_111_cast_fp16 = add(x = var_2975_cast_fp16, y = var_2979_cast_fp16)[name = string("state_111_cast_fp16")]; + tensor var_2982_begin_0 = const()[name = string("op_2982_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2982_end_0 = const()[name = string("op_2982_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2982_end_mask_0 = const()[name = string("op_2982_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2982_squeeze_mask_0 = const()[name = string("op_2982_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2982_cast_fp16 = slice_by_index(begin = var_2982_begin_0, end = var_2982_end_0, end_mask = var_2982_end_mask_0, squeeze_mask = var_2982_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_2982_cast_fp16")]; + tensor var_2984_begin_0 = const()[name = string("op_2984_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2984_end_0 = const()[name = string("op_2984_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2984_end_mask_0 = const()[name = string("op_2984_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2984_squeeze_mask_0 = const()[name = string("op_2984_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2984_cast_fp16 = slice_by_index(begin = var_2984_begin_0, end = var_2984_end_0, end_mask = var_2984_end_mask_0, squeeze_mask = var_2984_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_2984_cast_fp16")]; + bool var_2985_transpose_x_0 = const()[name = string("op_2985_transpose_x_0"), val = bool(false)]; + bool var_2985_transpose_y_0 = const()[name = string("op_2985_transpose_y_0"), val = bool(false)]; + tensor var_2985_cast_fp16 = matmul(transpose_x = var_2985_transpose_x_0, transpose_y = var_2985_transpose_y_0, x = var_2984_cast_fp16, y = state_111_cast_fp16)[name = string("op_2985_cast_fp16")]; + tensor v_new_111_cast_fp16 = sub(x = var_2982_cast_fp16, y = var_2985_cast_fp16)[name = string("v_new_111_cast_fp16")]; + tensor var_2988_begin_0 = const()[name = string("op_2988_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2988_end_0 = const()[name = string("op_2988_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2988_end_mask_0 = const()[name = string("op_2988_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2988_squeeze_mask_0 = const()[name = string("op_2988_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2988_cast_fp16 = slice_by_index(begin = var_2988_begin_0, end = var_2988_end_0, end_mask = var_2988_end_mask_0, squeeze_mask = var_2988_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_2988_cast_fp16")]; + bool var_2989_transpose_x_0 = const()[name = string("op_2989_transpose_x_0"), val = bool(false)]; + bool var_2989_transpose_y_0 = const()[name = string("op_2989_transpose_y_0"), val = bool(false)]; + tensor var_2989_cast_fp16 = matmul(transpose_x = var_2989_transpose_x_0, transpose_y = var_2989_transpose_y_0, x = var_2988_cast_fp16, y = state_111_cast_fp16)[name = string("op_2989_cast_fp16")]; + tensor var_2991_begin_0 = const()[name = string("op_2991_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2991_end_0 = const()[name = string("op_2991_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_2991_end_mask_0 = const()[name = string("op_2991_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2991_squeeze_mask_0 = const()[name = string("op_2991_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2991_cast_fp16 = slice_by_index(begin = var_2991_begin_0, end = var_2991_end_0, end_mask = var_2991_end_mask_0, squeeze_mask = var_2991_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_2991_cast_fp16")]; + bool var_2992_transpose_x_0 = const()[name = string("op_2992_transpose_x_0"), val = bool(false)]; + bool var_2992_transpose_y_0 = const()[name = string("op_2992_transpose_y_0"), val = bool(false)]; + tensor var_2992_cast_fp16 = matmul(transpose_x = var_2992_transpose_x_0, transpose_y = var_2992_transpose_y_0, x = var_2991_cast_fp16, y = v_new_111_cast_fp16)[name = string("op_2992_cast_fp16")]; + tensor var_2993_cast_fp16 = add(x = var_2989_cast_fp16, y = var_2992_cast_fp16)[name = string("op_2993_cast_fp16")]; + tensor var_2995_begin_0 = const()[name = string("op_2995_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2995_end_0 = const()[name = string("op_2995_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_2995_end_mask_0 = const()[name = string("op_2995_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2995_squeeze_mask_0 = const()[name = string("op_2995_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2995_cast_fp16 = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, squeeze_mask = var_2995_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_2995_cast_fp16")]; + tensor var_2996_cast_fp16 = mul(x = state_111_cast_fp16, y = var_2995_cast_fp16)[name = string("op_2996_cast_fp16")]; + tensor var_2998_begin_0 = const()[name = string("op_2998_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_2998_end_0 = const()[name = string("op_2998_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_2998_end_mask_0 = const()[name = string("op_2998_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2998_squeeze_mask_0 = const()[name = string("op_2998_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_2998_cast_fp16 = slice_by_index(begin = var_2998_begin_0, end = var_2998_end_0, end_mask = var_2998_end_mask_0, squeeze_mask = var_2998_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_2998_cast_fp16")]; + bool var_3000_transpose_x_1 = const()[name = string("op_3000_transpose_x_1"), val = bool(true)]; + bool var_3000_transpose_y_1 = const()[name = string("op_3000_transpose_y_1"), val = bool(false)]; + tensor var_3000_cast_fp16 = matmul(transpose_x = var_3000_transpose_x_1, transpose_y = var_3000_transpose_y_1, x = var_2998_cast_fp16, y = v_new_111_cast_fp16)[name = string("op_3000_cast_fp16")]; + tensor state_113_cast_fp16 = add(x = var_2996_cast_fp16, y = var_3000_cast_fp16)[name = string("state_113_cast_fp16")]; + tensor var_3003_begin_0 = const()[name = string("op_3003_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3003_end_0 = const()[name = string("op_3003_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3003_end_mask_0 = const()[name = string("op_3003_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3003_squeeze_mask_0 = const()[name = string("op_3003_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3003_cast_fp16 = slice_by_index(begin = var_3003_begin_0, end = var_3003_end_0, end_mask = var_3003_end_mask_0, squeeze_mask = var_3003_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3003_cast_fp16")]; + tensor var_3005_begin_0 = const()[name = string("op_3005_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3005_end_0 = const()[name = string("op_3005_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3005_end_mask_0 = const()[name = string("op_3005_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3005_squeeze_mask_0 = const()[name = string("op_3005_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3005_cast_fp16 = slice_by_index(begin = var_3005_begin_0, end = var_3005_end_0, end_mask = var_3005_end_mask_0, squeeze_mask = var_3005_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3005_cast_fp16")]; + bool var_3006_transpose_x_0 = const()[name = string("op_3006_transpose_x_0"), val = bool(false)]; + bool var_3006_transpose_y_0 = const()[name = string("op_3006_transpose_y_0"), val = bool(false)]; + tensor var_3006_cast_fp16 = matmul(transpose_x = var_3006_transpose_x_0, transpose_y = var_3006_transpose_y_0, x = var_3005_cast_fp16, y = state_113_cast_fp16)[name = string("op_3006_cast_fp16")]; + tensor v_new_113_cast_fp16 = sub(x = var_3003_cast_fp16, y = var_3006_cast_fp16)[name = string("v_new_113_cast_fp16")]; + tensor var_3009_begin_0 = const()[name = string("op_3009_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3009_end_0 = const()[name = string("op_3009_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3009_end_mask_0 = const()[name = string("op_3009_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3009_squeeze_mask_0 = const()[name = string("op_3009_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3009_cast_fp16 = slice_by_index(begin = var_3009_begin_0, end = var_3009_end_0, end_mask = var_3009_end_mask_0, squeeze_mask = var_3009_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3009_cast_fp16")]; + bool var_3010_transpose_x_0 = const()[name = string("op_3010_transpose_x_0"), val = bool(false)]; + bool var_3010_transpose_y_0 = const()[name = string("op_3010_transpose_y_0"), val = bool(false)]; + tensor var_3010_cast_fp16 = matmul(transpose_x = var_3010_transpose_x_0, transpose_y = var_3010_transpose_y_0, x = var_3009_cast_fp16, y = state_113_cast_fp16)[name = string("op_3010_cast_fp16")]; + tensor var_3012_begin_0 = const()[name = string("op_3012_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3012_end_0 = const()[name = string("op_3012_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_3012_end_mask_0 = const()[name = string("op_3012_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3012_squeeze_mask_0 = const()[name = string("op_3012_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3012_cast_fp16 = slice_by_index(begin = var_3012_begin_0, end = var_3012_end_0, end_mask = var_3012_end_mask_0, squeeze_mask = var_3012_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3012_cast_fp16")]; + bool var_3013_transpose_x_0 = const()[name = string("op_3013_transpose_x_0"), val = bool(false)]; + bool var_3013_transpose_y_0 = const()[name = string("op_3013_transpose_y_0"), val = bool(false)]; + tensor var_3013_cast_fp16 = matmul(transpose_x = var_3013_transpose_x_0, transpose_y = var_3013_transpose_y_0, x = var_3012_cast_fp16, y = v_new_113_cast_fp16)[name = string("op_3013_cast_fp16")]; + tensor var_3014_cast_fp16 = add(x = var_3010_cast_fp16, y = var_3013_cast_fp16)[name = string("op_3014_cast_fp16")]; + tensor var_3016_begin_0 = const()[name = string("op_3016_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3016_end_0 = const()[name = string("op_3016_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_3016_end_mask_0 = const()[name = string("op_3016_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3016_squeeze_mask_0 = const()[name = string("op_3016_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3016_cast_fp16 = slice_by_index(begin = var_3016_begin_0, end = var_3016_end_0, end_mask = var_3016_end_mask_0, squeeze_mask = var_3016_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3016_cast_fp16")]; + tensor var_3017_cast_fp16 = mul(x = state_113_cast_fp16, y = var_3016_cast_fp16)[name = string("op_3017_cast_fp16")]; + tensor var_3019_begin_0 = const()[name = string("op_3019_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3019_end_0 = const()[name = string("op_3019_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3019_end_mask_0 = const()[name = string("op_3019_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3019_squeeze_mask_0 = const()[name = string("op_3019_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3019_cast_fp16 = slice_by_index(begin = var_3019_begin_0, end = var_3019_end_0, end_mask = var_3019_end_mask_0, squeeze_mask = var_3019_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3019_cast_fp16")]; + bool var_3021_transpose_x_1 = const()[name = string("op_3021_transpose_x_1"), val = bool(true)]; + bool var_3021_transpose_y_1 = const()[name = string("op_3021_transpose_y_1"), val = bool(false)]; + tensor var_3021_cast_fp16 = matmul(transpose_x = var_3021_transpose_x_1, transpose_y = var_3021_transpose_y_1, x = var_3019_cast_fp16, y = v_new_113_cast_fp16)[name = string("op_3021_cast_fp16")]; + tensor state_115_cast_fp16 = add(x = var_3017_cast_fp16, y = var_3021_cast_fp16)[name = string("state_115_cast_fp16")]; + tensor var_3024_begin_0 = const()[name = string("op_3024_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3024_end_0 = const()[name = string("op_3024_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3024_end_mask_0 = const()[name = string("op_3024_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3024_squeeze_mask_0 = const()[name = string("op_3024_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3024_cast_fp16 = slice_by_index(begin = var_3024_begin_0, end = var_3024_end_0, end_mask = var_3024_end_mask_0, squeeze_mask = var_3024_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3024_cast_fp16")]; + tensor var_3026_begin_0 = const()[name = string("op_3026_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3026_end_0 = const()[name = string("op_3026_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3026_end_mask_0 = const()[name = string("op_3026_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3026_squeeze_mask_0 = const()[name = string("op_3026_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3026_cast_fp16 = slice_by_index(begin = var_3026_begin_0, end = var_3026_end_0, end_mask = var_3026_end_mask_0, squeeze_mask = var_3026_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3026_cast_fp16")]; + bool var_3027_transpose_x_0 = const()[name = string("op_3027_transpose_x_0"), val = bool(false)]; + bool var_3027_transpose_y_0 = const()[name = string("op_3027_transpose_y_0"), val = bool(false)]; + tensor var_3027_cast_fp16 = matmul(transpose_x = var_3027_transpose_x_0, transpose_y = var_3027_transpose_y_0, x = var_3026_cast_fp16, y = state_115_cast_fp16)[name = string("op_3027_cast_fp16")]; + tensor v_new_115_cast_fp16 = sub(x = var_3024_cast_fp16, y = var_3027_cast_fp16)[name = string("v_new_115_cast_fp16")]; + tensor var_3030_begin_0 = const()[name = string("op_3030_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3030_end_0 = const()[name = string("op_3030_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3030_end_mask_0 = const()[name = string("op_3030_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3030_squeeze_mask_0 = const()[name = string("op_3030_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3030_cast_fp16 = slice_by_index(begin = var_3030_begin_0, end = var_3030_end_0, end_mask = var_3030_end_mask_0, squeeze_mask = var_3030_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3030_cast_fp16")]; + bool var_3031_transpose_x_0 = const()[name = string("op_3031_transpose_x_0"), val = bool(false)]; + bool var_3031_transpose_y_0 = const()[name = string("op_3031_transpose_y_0"), val = bool(false)]; + tensor var_3031_cast_fp16 = matmul(transpose_x = var_3031_transpose_x_0, transpose_y = var_3031_transpose_y_0, x = var_3030_cast_fp16, y = state_115_cast_fp16)[name = string("op_3031_cast_fp16")]; + tensor var_3033_begin_0 = const()[name = string("op_3033_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3033_end_0 = const()[name = string("op_3033_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_3033_end_mask_0 = const()[name = string("op_3033_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3033_squeeze_mask_0 = const()[name = string("op_3033_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3033_cast_fp16 = slice_by_index(begin = var_3033_begin_0, end = var_3033_end_0, end_mask = var_3033_end_mask_0, squeeze_mask = var_3033_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3033_cast_fp16")]; + bool var_3034_transpose_x_0 = const()[name = string("op_3034_transpose_x_0"), val = bool(false)]; + bool var_3034_transpose_y_0 = const()[name = string("op_3034_transpose_y_0"), val = bool(false)]; + tensor var_3034_cast_fp16 = matmul(transpose_x = var_3034_transpose_x_0, transpose_y = var_3034_transpose_y_0, x = var_3033_cast_fp16, y = v_new_115_cast_fp16)[name = string("op_3034_cast_fp16")]; + tensor var_3035_cast_fp16 = add(x = var_3031_cast_fp16, y = var_3034_cast_fp16)[name = string("op_3035_cast_fp16")]; + tensor var_3037_begin_0 = const()[name = string("op_3037_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3037_end_0 = const()[name = string("op_3037_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_3037_end_mask_0 = const()[name = string("op_3037_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3037_squeeze_mask_0 = const()[name = string("op_3037_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3037_cast_fp16 = slice_by_index(begin = var_3037_begin_0, end = var_3037_end_0, end_mask = var_3037_end_mask_0, squeeze_mask = var_3037_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3037_cast_fp16")]; + tensor var_3038_cast_fp16 = mul(x = state_115_cast_fp16, y = var_3037_cast_fp16)[name = string("op_3038_cast_fp16")]; + tensor var_3040_begin_0 = const()[name = string("op_3040_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3040_end_0 = const()[name = string("op_3040_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3040_end_mask_0 = const()[name = string("op_3040_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3040_squeeze_mask_0 = const()[name = string("op_3040_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3040_cast_fp16 = slice_by_index(begin = var_3040_begin_0, end = var_3040_end_0, end_mask = var_3040_end_mask_0, squeeze_mask = var_3040_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3040_cast_fp16")]; + bool var_3042_transpose_x_1 = const()[name = string("op_3042_transpose_x_1"), val = bool(true)]; + bool var_3042_transpose_y_1 = const()[name = string("op_3042_transpose_y_1"), val = bool(false)]; + tensor var_3042_cast_fp16 = matmul(transpose_x = var_3042_transpose_x_1, transpose_y = var_3042_transpose_y_1, x = var_3040_cast_fp16, y = v_new_115_cast_fp16)[name = string("op_3042_cast_fp16")]; + tensor state_117_cast_fp16 = add(x = var_3038_cast_fp16, y = var_3042_cast_fp16)[name = string("state_117_cast_fp16")]; + tensor var_3045_begin_0 = const()[name = string("op_3045_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3045_end_0 = const()[name = string("op_3045_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3045_end_mask_0 = const()[name = string("op_3045_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3045_squeeze_mask_0 = const()[name = string("op_3045_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3045_cast_fp16 = slice_by_index(begin = var_3045_begin_0, end = var_3045_end_0, end_mask = var_3045_end_mask_0, squeeze_mask = var_3045_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3045_cast_fp16")]; + tensor var_3047_begin_0 = const()[name = string("op_3047_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3047_end_0 = const()[name = string("op_3047_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3047_end_mask_0 = const()[name = string("op_3047_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3047_squeeze_mask_0 = const()[name = string("op_3047_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3047_cast_fp16 = slice_by_index(begin = var_3047_begin_0, end = var_3047_end_0, end_mask = var_3047_end_mask_0, squeeze_mask = var_3047_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3047_cast_fp16")]; + bool var_3048_transpose_x_0 = const()[name = string("op_3048_transpose_x_0"), val = bool(false)]; + bool var_3048_transpose_y_0 = const()[name = string("op_3048_transpose_y_0"), val = bool(false)]; + tensor var_3048_cast_fp16 = matmul(transpose_x = var_3048_transpose_x_0, transpose_y = var_3048_transpose_y_0, x = var_3047_cast_fp16, y = state_117_cast_fp16)[name = string("op_3048_cast_fp16")]; + tensor v_new_117_cast_fp16 = sub(x = var_3045_cast_fp16, y = var_3048_cast_fp16)[name = string("v_new_117_cast_fp16")]; + tensor var_3051_begin_0 = const()[name = string("op_3051_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3051_end_0 = const()[name = string("op_3051_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3051_end_mask_0 = const()[name = string("op_3051_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3051_squeeze_mask_0 = const()[name = string("op_3051_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3051_cast_fp16 = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, squeeze_mask = var_3051_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3051_cast_fp16")]; + bool var_3052_transpose_x_0 = const()[name = string("op_3052_transpose_x_0"), val = bool(false)]; + bool var_3052_transpose_y_0 = const()[name = string("op_3052_transpose_y_0"), val = bool(false)]; + tensor var_3052_cast_fp16 = matmul(transpose_x = var_3052_transpose_x_0, transpose_y = var_3052_transpose_y_0, x = var_3051_cast_fp16, y = state_117_cast_fp16)[name = string("op_3052_cast_fp16")]; + tensor var_3054_begin_0 = const()[name = string("op_3054_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3054_end_0 = const()[name = string("op_3054_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_3054_end_mask_0 = const()[name = string("op_3054_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3054_squeeze_mask_0 = const()[name = string("op_3054_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3054_cast_fp16 = slice_by_index(begin = var_3054_begin_0, end = var_3054_end_0, end_mask = var_3054_end_mask_0, squeeze_mask = var_3054_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3054_cast_fp16")]; + bool var_3055_transpose_x_0 = const()[name = string("op_3055_transpose_x_0"), val = bool(false)]; + bool var_3055_transpose_y_0 = const()[name = string("op_3055_transpose_y_0"), val = bool(false)]; + tensor var_3055_cast_fp16 = matmul(transpose_x = var_3055_transpose_x_0, transpose_y = var_3055_transpose_y_0, x = var_3054_cast_fp16, y = v_new_117_cast_fp16)[name = string("op_3055_cast_fp16")]; + tensor var_3056_cast_fp16 = add(x = var_3052_cast_fp16, y = var_3055_cast_fp16)[name = string("op_3056_cast_fp16")]; + tensor var_3058_begin_0 = const()[name = string("op_3058_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3058_end_0 = const()[name = string("op_3058_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_3058_end_mask_0 = const()[name = string("op_3058_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3058_squeeze_mask_0 = const()[name = string("op_3058_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3058_cast_fp16 = slice_by_index(begin = var_3058_begin_0, end = var_3058_end_0, end_mask = var_3058_end_mask_0, squeeze_mask = var_3058_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3058_cast_fp16")]; + tensor var_3059_cast_fp16 = mul(x = state_117_cast_fp16, y = var_3058_cast_fp16)[name = string("op_3059_cast_fp16")]; + tensor var_3061_begin_0 = const()[name = string("op_3061_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3061_end_0 = const()[name = string("op_3061_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3061_end_mask_0 = const()[name = string("op_3061_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3061_squeeze_mask_0 = const()[name = string("op_3061_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3061_cast_fp16 = slice_by_index(begin = var_3061_begin_0, end = var_3061_end_0, end_mask = var_3061_end_mask_0, squeeze_mask = var_3061_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3061_cast_fp16")]; + bool var_3063_transpose_x_1 = const()[name = string("op_3063_transpose_x_1"), val = bool(true)]; + bool var_3063_transpose_y_1 = const()[name = string("op_3063_transpose_y_1"), val = bool(false)]; + tensor var_3063_cast_fp16 = matmul(transpose_x = var_3063_transpose_x_1, transpose_y = var_3063_transpose_y_1, x = var_3061_cast_fp16, y = v_new_117_cast_fp16)[name = string("op_3063_cast_fp16")]; + tensor state_119_cast_fp16 = add(x = var_3059_cast_fp16, y = var_3063_cast_fp16)[name = string("state_119_cast_fp16")]; + tensor var_3066_begin_0 = const()[name = string("op_3066_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3066_end_0 = const()[name = string("op_3066_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3066_end_mask_0 = const()[name = string("op_3066_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3066_squeeze_mask_0 = const()[name = string("op_3066_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3066_cast_fp16 = slice_by_index(begin = var_3066_begin_0, end = var_3066_end_0, end_mask = var_3066_end_mask_0, squeeze_mask = var_3066_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3066_cast_fp16")]; + tensor var_3068_begin_0 = const()[name = string("op_3068_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3068_end_0 = const()[name = string("op_3068_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3068_end_mask_0 = const()[name = string("op_3068_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3068_squeeze_mask_0 = const()[name = string("op_3068_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3068_cast_fp16 = slice_by_index(begin = var_3068_begin_0, end = var_3068_end_0, end_mask = var_3068_end_mask_0, squeeze_mask = var_3068_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3068_cast_fp16")]; + bool var_3069_transpose_x_0 = const()[name = string("op_3069_transpose_x_0"), val = bool(false)]; + bool var_3069_transpose_y_0 = const()[name = string("op_3069_transpose_y_0"), val = bool(false)]; + tensor var_3069_cast_fp16 = matmul(transpose_x = var_3069_transpose_x_0, transpose_y = var_3069_transpose_y_0, x = var_3068_cast_fp16, y = state_119_cast_fp16)[name = string("op_3069_cast_fp16")]; + tensor v_new_119_cast_fp16 = sub(x = var_3066_cast_fp16, y = var_3069_cast_fp16)[name = string("v_new_119_cast_fp16")]; + tensor var_3072_begin_0 = const()[name = string("op_3072_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3072_end_0 = const()[name = string("op_3072_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3072_end_mask_0 = const()[name = string("op_3072_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3072_squeeze_mask_0 = const()[name = string("op_3072_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3072_cast_fp16 = slice_by_index(begin = var_3072_begin_0, end = var_3072_end_0, end_mask = var_3072_end_mask_0, squeeze_mask = var_3072_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3072_cast_fp16")]; + bool var_3073_transpose_x_0 = const()[name = string("op_3073_transpose_x_0"), val = bool(false)]; + bool var_3073_transpose_y_0 = const()[name = string("op_3073_transpose_y_0"), val = bool(false)]; + tensor var_3073_cast_fp16 = matmul(transpose_x = var_3073_transpose_x_0, transpose_y = var_3073_transpose_y_0, x = var_3072_cast_fp16, y = state_119_cast_fp16)[name = string("op_3073_cast_fp16")]; + tensor var_3075_begin_0 = const()[name = string("op_3075_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3075_end_0 = const()[name = string("op_3075_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_3075_end_mask_0 = const()[name = string("op_3075_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3075_squeeze_mask_0 = const()[name = string("op_3075_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3075_cast_fp16 = slice_by_index(begin = var_3075_begin_0, end = var_3075_end_0, end_mask = var_3075_end_mask_0, squeeze_mask = var_3075_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3075_cast_fp16")]; + bool var_3076_transpose_x_0 = const()[name = string("op_3076_transpose_x_0"), val = bool(false)]; + bool var_3076_transpose_y_0 = const()[name = string("op_3076_transpose_y_0"), val = bool(false)]; + tensor var_3076_cast_fp16 = matmul(transpose_x = var_3076_transpose_x_0, transpose_y = var_3076_transpose_y_0, x = var_3075_cast_fp16, y = v_new_119_cast_fp16)[name = string("op_3076_cast_fp16")]; + tensor var_3077_cast_fp16 = add(x = var_3073_cast_fp16, y = var_3076_cast_fp16)[name = string("op_3077_cast_fp16")]; + tensor var_3079_begin_0 = const()[name = string("op_3079_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3079_end_0 = const()[name = string("op_3079_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_3079_end_mask_0 = const()[name = string("op_3079_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3079_squeeze_mask_0 = const()[name = string("op_3079_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3079_cast_fp16 = slice_by_index(begin = var_3079_begin_0, end = var_3079_end_0, end_mask = var_3079_end_mask_0, squeeze_mask = var_3079_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3079_cast_fp16")]; + tensor var_3080_cast_fp16 = mul(x = state_119_cast_fp16, y = var_3079_cast_fp16)[name = string("op_3080_cast_fp16")]; + tensor var_3082_begin_0 = const()[name = string("op_3082_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3082_end_0 = const()[name = string("op_3082_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3082_end_mask_0 = const()[name = string("op_3082_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3082_squeeze_mask_0 = const()[name = string("op_3082_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3082_cast_fp16 = slice_by_index(begin = var_3082_begin_0, end = var_3082_end_0, end_mask = var_3082_end_mask_0, squeeze_mask = var_3082_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3082_cast_fp16")]; + bool var_3084_transpose_x_1 = const()[name = string("op_3084_transpose_x_1"), val = bool(true)]; + bool var_3084_transpose_y_1 = const()[name = string("op_3084_transpose_y_1"), val = bool(false)]; + tensor var_3084_cast_fp16 = matmul(transpose_x = var_3084_transpose_x_1, transpose_y = var_3084_transpose_y_1, x = var_3082_cast_fp16, y = v_new_119_cast_fp16)[name = string("op_3084_cast_fp16")]; + tensor state_121_cast_fp16 = add(x = var_3080_cast_fp16, y = var_3084_cast_fp16)[name = string("state_121_cast_fp16")]; + tensor var_3087_begin_0 = const()[name = string("op_3087_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3087_end_0 = const()[name = string("op_3087_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3087_end_mask_0 = const()[name = string("op_3087_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3087_squeeze_mask_0 = const()[name = string("op_3087_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3087_cast_fp16 = slice_by_index(begin = var_3087_begin_0, end = var_3087_end_0, end_mask = var_3087_end_mask_0, squeeze_mask = var_3087_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3087_cast_fp16")]; + tensor var_3089_begin_0 = const()[name = string("op_3089_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3089_end_0 = const()[name = string("op_3089_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3089_end_mask_0 = const()[name = string("op_3089_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3089_squeeze_mask_0 = const()[name = string("op_3089_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3089_cast_fp16 = slice_by_index(begin = var_3089_begin_0, end = var_3089_end_0, end_mask = var_3089_end_mask_0, squeeze_mask = var_3089_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3089_cast_fp16")]; + bool var_3090_transpose_x_0 = const()[name = string("op_3090_transpose_x_0"), val = bool(false)]; + bool var_3090_transpose_y_0 = const()[name = string("op_3090_transpose_y_0"), val = bool(false)]; + tensor var_3090_cast_fp16 = matmul(transpose_x = var_3090_transpose_x_0, transpose_y = var_3090_transpose_y_0, x = var_3089_cast_fp16, y = state_121_cast_fp16)[name = string("op_3090_cast_fp16")]; + tensor v_new_121_cast_fp16 = sub(x = var_3087_cast_fp16, y = var_3090_cast_fp16)[name = string("v_new_121_cast_fp16")]; + tensor var_3093_begin_0 = const()[name = string("op_3093_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3093_end_0 = const()[name = string("op_3093_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3093_end_mask_0 = const()[name = string("op_3093_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3093_squeeze_mask_0 = const()[name = string("op_3093_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3093_cast_fp16 = slice_by_index(begin = var_3093_begin_0, end = var_3093_end_0, end_mask = var_3093_end_mask_0, squeeze_mask = var_3093_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3093_cast_fp16")]; + bool var_3094_transpose_x_0 = const()[name = string("op_3094_transpose_x_0"), val = bool(false)]; + bool var_3094_transpose_y_0 = const()[name = string("op_3094_transpose_y_0"), val = bool(false)]; + tensor var_3094_cast_fp16 = matmul(transpose_x = var_3094_transpose_x_0, transpose_y = var_3094_transpose_y_0, x = var_3093_cast_fp16, y = state_121_cast_fp16)[name = string("op_3094_cast_fp16")]; + tensor var_3096_begin_0 = const()[name = string("op_3096_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3096_end_0 = const()[name = string("op_3096_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_3096_end_mask_0 = const()[name = string("op_3096_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3096_squeeze_mask_0 = const()[name = string("op_3096_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3096_cast_fp16 = slice_by_index(begin = var_3096_begin_0, end = var_3096_end_0, end_mask = var_3096_end_mask_0, squeeze_mask = var_3096_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3096_cast_fp16")]; + bool var_3097_transpose_x_0 = const()[name = string("op_3097_transpose_x_0"), val = bool(false)]; + bool var_3097_transpose_y_0 = const()[name = string("op_3097_transpose_y_0"), val = bool(false)]; + tensor var_3097_cast_fp16 = matmul(transpose_x = var_3097_transpose_x_0, transpose_y = var_3097_transpose_y_0, x = var_3096_cast_fp16, y = v_new_121_cast_fp16)[name = string("op_3097_cast_fp16")]; + tensor var_3098_cast_fp16 = add(x = var_3094_cast_fp16, y = var_3097_cast_fp16)[name = string("op_3098_cast_fp16")]; + tensor var_3100_begin_0 = const()[name = string("op_3100_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3100_end_0 = const()[name = string("op_3100_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_3100_end_mask_0 = const()[name = string("op_3100_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3100_squeeze_mask_0 = const()[name = string("op_3100_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3100_cast_fp16 = slice_by_index(begin = var_3100_begin_0, end = var_3100_end_0, end_mask = var_3100_end_mask_0, squeeze_mask = var_3100_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3100_cast_fp16")]; + tensor var_3101_cast_fp16 = mul(x = state_121_cast_fp16, y = var_3100_cast_fp16)[name = string("op_3101_cast_fp16")]; + tensor var_3103_begin_0 = const()[name = string("op_3103_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3103_end_0 = const()[name = string("op_3103_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3103_end_mask_0 = const()[name = string("op_3103_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3103_squeeze_mask_0 = const()[name = string("op_3103_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3103_cast_fp16 = slice_by_index(begin = var_3103_begin_0, end = var_3103_end_0, end_mask = var_3103_end_mask_0, squeeze_mask = var_3103_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3103_cast_fp16")]; + bool var_3105_transpose_x_1 = const()[name = string("op_3105_transpose_x_1"), val = bool(true)]; + bool var_3105_transpose_y_1 = const()[name = string("op_3105_transpose_y_1"), val = bool(false)]; + tensor var_3105_cast_fp16 = matmul(transpose_x = var_3105_transpose_x_1, transpose_y = var_3105_transpose_y_1, x = var_3103_cast_fp16, y = v_new_121_cast_fp16)[name = string("op_3105_cast_fp16")]; + tensor state_123_cast_fp16 = add(x = var_3101_cast_fp16, y = var_3105_cast_fp16)[name = string("state_123_cast_fp16")]; + tensor var_3108_begin_0 = const()[name = string("op_3108_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3108_end_0 = const()[name = string("op_3108_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3108_end_mask_0 = const()[name = string("op_3108_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3108_squeeze_mask_0 = const()[name = string("op_3108_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3108_cast_fp16 = slice_by_index(begin = var_3108_begin_0, end = var_3108_end_0, end_mask = var_3108_end_mask_0, squeeze_mask = var_3108_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3108_cast_fp16")]; + tensor var_3110_begin_0 = const()[name = string("op_3110_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3110_end_0 = const()[name = string("op_3110_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3110_end_mask_0 = const()[name = string("op_3110_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3110_squeeze_mask_0 = const()[name = string("op_3110_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3110_cast_fp16 = slice_by_index(begin = var_3110_begin_0, end = var_3110_end_0, end_mask = var_3110_end_mask_0, squeeze_mask = var_3110_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3110_cast_fp16")]; + bool var_3111_transpose_x_0 = const()[name = string("op_3111_transpose_x_0"), val = bool(false)]; + bool var_3111_transpose_y_0 = const()[name = string("op_3111_transpose_y_0"), val = bool(false)]; + tensor var_3111_cast_fp16 = matmul(transpose_x = var_3111_transpose_x_0, transpose_y = var_3111_transpose_y_0, x = var_3110_cast_fp16, y = state_123_cast_fp16)[name = string("op_3111_cast_fp16")]; + tensor v_new_123_cast_fp16 = sub(x = var_3108_cast_fp16, y = var_3111_cast_fp16)[name = string("v_new_123_cast_fp16")]; + tensor var_3114_begin_0 = const()[name = string("op_3114_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3114_end_0 = const()[name = string("op_3114_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3114_end_mask_0 = const()[name = string("op_3114_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3114_squeeze_mask_0 = const()[name = string("op_3114_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3114_cast_fp16 = slice_by_index(begin = var_3114_begin_0, end = var_3114_end_0, end_mask = var_3114_end_mask_0, squeeze_mask = var_3114_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3114_cast_fp16")]; + bool var_3115_transpose_x_0 = const()[name = string("op_3115_transpose_x_0"), val = bool(false)]; + bool var_3115_transpose_y_0 = const()[name = string("op_3115_transpose_y_0"), val = bool(false)]; + tensor var_3115_cast_fp16 = matmul(transpose_x = var_3115_transpose_x_0, transpose_y = var_3115_transpose_y_0, x = var_3114_cast_fp16, y = state_123_cast_fp16)[name = string("op_3115_cast_fp16")]; + tensor var_3117_begin_0 = const()[name = string("op_3117_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3117_end_0 = const()[name = string("op_3117_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_3117_end_mask_0 = const()[name = string("op_3117_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3117_squeeze_mask_0 = const()[name = string("op_3117_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3117_cast_fp16 = slice_by_index(begin = var_3117_begin_0, end = var_3117_end_0, end_mask = var_3117_end_mask_0, squeeze_mask = var_3117_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3117_cast_fp16")]; + bool var_3118_transpose_x_0 = const()[name = string("op_3118_transpose_x_0"), val = bool(false)]; + bool var_3118_transpose_y_0 = const()[name = string("op_3118_transpose_y_0"), val = bool(false)]; + tensor var_3118_cast_fp16 = matmul(transpose_x = var_3118_transpose_x_0, transpose_y = var_3118_transpose_y_0, x = var_3117_cast_fp16, y = v_new_123_cast_fp16)[name = string("op_3118_cast_fp16")]; + tensor var_3119_cast_fp16 = add(x = var_3115_cast_fp16, y = var_3118_cast_fp16)[name = string("op_3119_cast_fp16")]; + tensor var_3121_begin_0 = const()[name = string("op_3121_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3121_end_0 = const()[name = string("op_3121_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_3121_end_mask_0 = const()[name = string("op_3121_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3121_squeeze_mask_0 = const()[name = string("op_3121_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3121_cast_fp16 = slice_by_index(begin = var_3121_begin_0, end = var_3121_end_0, end_mask = var_3121_end_mask_0, squeeze_mask = var_3121_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3121_cast_fp16")]; + tensor var_3122_cast_fp16 = mul(x = state_123_cast_fp16, y = var_3121_cast_fp16)[name = string("op_3122_cast_fp16")]; + tensor var_3124_begin_0 = const()[name = string("op_3124_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3124_end_0 = const()[name = string("op_3124_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3124_end_mask_0 = const()[name = string("op_3124_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3124_squeeze_mask_0 = const()[name = string("op_3124_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3124_cast_fp16 = slice_by_index(begin = var_3124_begin_0, end = var_3124_end_0, end_mask = var_3124_end_mask_0, squeeze_mask = var_3124_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3124_cast_fp16")]; + bool var_3126_transpose_x_1 = const()[name = string("op_3126_transpose_x_1"), val = bool(true)]; + bool var_3126_transpose_y_1 = const()[name = string("op_3126_transpose_y_1"), val = bool(false)]; + tensor var_3126_cast_fp16 = matmul(transpose_x = var_3126_transpose_x_1, transpose_y = var_3126_transpose_y_1, x = var_3124_cast_fp16, y = v_new_123_cast_fp16)[name = string("op_3126_cast_fp16")]; + tensor state_125_cast_fp16 = add(x = var_3122_cast_fp16, y = var_3126_cast_fp16)[name = string("state_125_cast_fp16")]; + tensor var_3129_begin_0 = const()[name = string("op_3129_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3129_end_0 = const()[name = string("op_3129_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3129_end_mask_0 = const()[name = string("op_3129_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3129_squeeze_mask_0 = const()[name = string("op_3129_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3129_cast_fp16 = slice_by_index(begin = var_3129_begin_0, end = var_3129_end_0, end_mask = var_3129_end_mask_0, squeeze_mask = var_3129_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3129_cast_fp16")]; + tensor var_3131_begin_0 = const()[name = string("op_3131_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3131_end_0 = const()[name = string("op_3131_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3131_end_mask_0 = const()[name = string("op_3131_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3131_squeeze_mask_0 = const()[name = string("op_3131_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3131_cast_fp16 = slice_by_index(begin = var_3131_begin_0, end = var_3131_end_0, end_mask = var_3131_end_mask_0, squeeze_mask = var_3131_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3131_cast_fp16")]; + bool var_3132_transpose_x_0 = const()[name = string("op_3132_transpose_x_0"), val = bool(false)]; + bool var_3132_transpose_y_0 = const()[name = string("op_3132_transpose_y_0"), val = bool(false)]; + tensor var_3132_cast_fp16 = matmul(transpose_x = var_3132_transpose_x_0, transpose_y = var_3132_transpose_y_0, x = var_3131_cast_fp16, y = state_125_cast_fp16)[name = string("op_3132_cast_fp16")]; + tensor v_new_125_cast_fp16 = sub(x = var_3129_cast_fp16, y = var_3132_cast_fp16)[name = string("v_new_125_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3135_squeeze_mask_0 = const()[name = string("op_3135_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, squeeze_mask = var_3135_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3135_cast_fp16")]; + bool var_3136_transpose_x_0 = const()[name = string("op_3136_transpose_x_0"), val = bool(false)]; + bool var_3136_transpose_y_0 = const()[name = string("op_3136_transpose_y_0"), val = bool(false)]; + tensor var_3136_cast_fp16 = matmul(transpose_x = var_3136_transpose_x_0, transpose_y = var_3136_transpose_y_0, x = var_3135_cast_fp16, y = state_125_cast_fp16)[name = string("op_3136_cast_fp16")]; + tensor var_3138_begin_0 = const()[name = string("op_3138_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3138_end_0 = const()[name = string("op_3138_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_3138_end_mask_0 = const()[name = string("op_3138_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3138_squeeze_mask_0 = const()[name = string("op_3138_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3138_cast_fp16 = slice_by_index(begin = var_3138_begin_0, end = var_3138_end_0, end_mask = var_3138_end_mask_0, squeeze_mask = var_3138_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3138_cast_fp16")]; + bool var_3139_transpose_x_0 = const()[name = string("op_3139_transpose_x_0"), val = bool(false)]; + bool var_3139_transpose_y_0 = const()[name = string("op_3139_transpose_y_0"), val = bool(false)]; + tensor var_3139_cast_fp16 = matmul(transpose_x = var_3139_transpose_x_0, transpose_y = var_3139_transpose_y_0, x = var_3138_cast_fp16, y = v_new_125_cast_fp16)[name = string("op_3139_cast_fp16")]; + tensor var_3140_cast_fp16 = add(x = var_3136_cast_fp16, y = var_3139_cast_fp16)[name = string("op_3140_cast_fp16")]; + tensor var_3142_begin_0 = const()[name = string("op_3142_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3142_end_0 = const()[name = string("op_3142_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_3142_end_mask_0 = const()[name = string("op_3142_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3142_squeeze_mask_0 = const()[name = string("op_3142_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3142_cast_fp16 = slice_by_index(begin = var_3142_begin_0, end = var_3142_end_0, end_mask = var_3142_end_mask_0, squeeze_mask = var_3142_squeeze_mask_0, x = chunk_decay_7_cast_fp16)[name = string("op_3142_cast_fp16")]; + tensor var_3143_cast_fp16 = mul(x = state_125_cast_fp16, y = var_3142_cast_fp16)[name = string("op_3143_cast_fp16")]; + tensor var_3145_begin_0 = const()[name = string("op_3145_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3145_end_0 = const()[name = string("op_3145_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3145_end_mask_0 = const()[name = string("op_3145_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3145_squeeze_mask_0 = const()[name = string("op_3145_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3145_cast_fp16 = slice_by_index(begin = var_3145_begin_0, end = var_3145_end_0, end_mask = var_3145_end_mask_0, squeeze_mask = var_3145_squeeze_mask_0, x = k_dec_7_cast_fp16)[name = string("op_3145_cast_fp16")]; + bool var_3147_transpose_x_1 = const()[name = string("op_3147_transpose_x_1"), val = bool(true)]; + bool var_3147_transpose_y_1 = const()[name = string("op_3147_transpose_y_1"), val = bool(false)]; + tensor var_3147_cast_fp16 = matmul(transpose_x = var_3147_transpose_x_1, transpose_y = var_3147_transpose_y_1, x = var_3145_cast_fp16, y = v_new_125_cast_fp16)[name = string("op_3147_cast_fp16")]; + tensor state_127_cast_fp16 = add(x = var_3143_cast_fp16, y = var_3147_cast_fp16)[name = string("state_127_cast_fp16")]; + tensor var_3150_begin_0 = const()[name = string("op_3150_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3150_end_0 = const()[name = string("op_3150_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3150_end_mask_0 = const()[name = string("op_3150_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3150_squeeze_mask_0 = const()[name = string("op_3150_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3150_cast_fp16 = slice_by_index(begin = var_3150_begin_0, end = var_3150_end_0, end_mask = var_3150_end_mask_0, squeeze_mask = var_3150_squeeze_mask_0, x = new_v_7_cast_fp16)[name = string("op_3150_cast_fp16")]; + tensor var_3152_begin_0 = const()[name = string("op_3152_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3152_end_0 = const()[name = string("op_3152_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3152_end_mask_0 = const()[name = string("op_3152_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3152_squeeze_mask_0 = const()[name = string("op_3152_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3152_cast_fp16 = slice_by_index(begin = var_3152_begin_0, end = var_3152_end_0, end_mask = var_3152_end_mask_0, squeeze_mask = var_3152_squeeze_mask_0, x = k_cumdecay_7_cast_fp16)[name = string("op_3152_cast_fp16")]; + bool var_3153_transpose_x_0 = const()[name = string("op_3153_transpose_x_0"), val = bool(false)]; + bool var_3153_transpose_y_0 = const()[name = string("op_3153_transpose_y_0"), val = bool(false)]; + tensor var_3153_cast_fp16 = matmul(transpose_x = var_3153_transpose_x_0, transpose_y = var_3153_transpose_y_0, x = var_3152_cast_fp16, y = state_127_cast_fp16)[name = string("op_3153_cast_fp16")]; + tensor v_new_127_cast_fp16 = sub(x = var_3150_cast_fp16, y = var_3153_cast_fp16)[name = string("v_new_127_cast_fp16")]; + tensor var_3156_begin_0 = const()[name = string("op_3156_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3156_end_0 = const()[name = string("op_3156_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3156_end_mask_0 = const()[name = string("op_3156_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3156_squeeze_mask_0 = const()[name = string("op_3156_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3156_cast_fp16 = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, squeeze_mask = var_3156_squeeze_mask_0, x = q_dec_7_cast_fp16)[name = string("op_3156_cast_fp16")]; + bool var_3157_transpose_x_0 = const()[name = string("op_3157_transpose_x_0"), val = bool(false)]; + bool var_3157_transpose_y_0 = const()[name = string("op_3157_transpose_y_0"), val = bool(false)]; + tensor var_3157_cast_fp16 = matmul(transpose_x = var_3157_transpose_x_0, transpose_y = var_3157_transpose_y_0, x = var_3156_cast_fp16, y = state_127_cast_fp16)[name = string("op_3157_cast_fp16")]; + tensor var_3159_begin_0 = const()[name = string("op_3159_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3159_end_0 = const()[name = string("op_3159_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_3159_end_mask_0 = const()[name = string("op_3159_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3159_squeeze_mask_0 = const()[name = string("op_3159_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3159_cast_fp16 = slice_by_index(begin = var_3159_begin_0, end = var_3159_end_0, end_mask = var_3159_end_mask_0, squeeze_mask = var_3159_squeeze_mask_0, x = attn_intra_7_cast_fp16)[name = string("op_3159_cast_fp16")]; + bool var_3160_transpose_x_0 = const()[name = string("op_3160_transpose_x_0"), val = bool(false)]; + bool var_3160_transpose_y_0 = const()[name = string("op_3160_transpose_y_0"), val = bool(false)]; + tensor var_3160_cast_fp16 = matmul(transpose_x = var_3160_transpose_x_0, transpose_y = var_3160_transpose_y_0, x = var_3159_cast_fp16, y = v_new_127_cast_fp16)[name = string("op_3160_cast_fp16")]; + tensor var_3161_cast_fp16 = add(x = var_3157_cast_fp16, y = var_3160_cast_fp16)[name = string("op_3161_cast_fp16")]; + int32 var_3163_axis_0 = const()[name = string("op_3163_axis_0"), val = int32(1)]; + tensor var_3163_cast_fp16 = stack(axis = var_3163_axis_0, values = (var_2846_cast_fp16, var_2867_cast_fp16, var_2888_cast_fp16, var_2909_cast_fp16, var_2930_cast_fp16, var_2951_cast_fp16, var_2972_cast_fp16, var_2993_cast_fp16, var_3014_cast_fp16, var_3035_cast_fp16, var_3056_cast_fp16, var_3077_cast_fp16, var_3098_cast_fp16, var_3119_cast_fp16, var_3140_cast_fp16, var_3161_cast_fp16))[name = string("op_3163_cast_fp16")]; + tensor var_3164 = const()[name = string("op_3164"), val = tensor([32, 1024, 128])]; + tensor var_3165_cast_fp16 = reshape(shape = var_3164, x = var_3163_cast_fp16)[name = string("op_3165_cast_fp16")]; + tensor var_3166 = const()[name = string("op_3166"), val = tensor([1, 0, 2])]; + tensor input_63_begin_0 = const()[name = string("input_63_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_63_end_0 = const()[name = string("input_63_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_63_end_mask_0 = const()[name = string("input_63_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_63_squeeze_mask_0 = const()[name = string("input_63_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_63_cast_fp16 = slice_by_index(begin = input_63_begin_0, end = input_63_end_0, end_mask = input_63_end_mask_0, squeeze_mask = input_63_squeeze_mask_0, x = z_7_cast_fp16)[name = string("input_63_cast_fp16")]; + fp16 var_2488_promoted_1_to_fp16 = const()[name = string("op_2488_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_117_cast_fp16 = transpose(perm = var_3166, x = var_3165_cast_fp16)[name = string("transpose_20")]; + tensor var_3170_cast_fp16 = pow(x = x_117_cast_fp16, y = var_2488_promoted_1_to_fp16)[name = string("op_3170_cast_fp16")]; + tensor var_3172_axes_0 = const()[name = string("op_3172_axes_0"), val = tensor([-1])]; + bool var_3172_keep_dims_0 = const()[name = string("op_3172_keep_dims_0"), val = bool(true)]; + tensor var_3172_cast_fp16 = reduce_mean(axes = var_3172_axes_0, keep_dims = var_3172_keep_dims_0, x = var_3170_cast_fp16)[name = string("op_3172_cast_fp16")]; + fp16 var_3173_to_fp16 = const()[name = string("op_3173_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3174_cast_fp16 = add(x = var_3172_cast_fp16, y = var_3173_to_fp16)[name = string("op_3174_cast_fp16")]; + fp32 var_3175_epsilon_0 = const()[name = string("op_3175_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3175_cast_fp16 = rsqrt(epsilon = var_3175_epsilon_0, x = var_3174_cast_fp16)[name = string("op_3175_cast_fp16")]; + tensor x_119_cast_fp16 = mul(x = x_117_cast_fp16, y = var_3175_cast_fp16)[name = string("x_119_cast_fp16")]; + tensor layers_4_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_4_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961717952)))]; + tensor var_3177_cast_fp16 = mul(x = layers_4_linear_attn_norm_weight_to_fp16, y = x_119_cast_fp16)[name = string("op_3177_cast_fp16")]; + tensor var_3178_cast_fp16 = silu(x = input_63_cast_fp16)[name = string("op_3178_cast_fp16")]; + tensor core_7_cast_fp16 = mul(x = var_3177_cast_fp16, y = var_3178_cast_fp16)[name = string("core_7_cast_fp16")]; + tensor var_3180 = const()[name = string("op_3180"), val = tensor([1, 1024, 4096])]; + tensor input_65_cast_fp16 = reshape(shape = var_3180, x = core_7_cast_fp16)[name = string("input_65_cast_fp16")]; + tensor layers_4_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_4_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(961718272)))]; + tensor linear_35_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_4_linear_attn_out_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_35_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_99_cast_fp16, y = linear_35_cast_fp16)[name = string("x_121_cast_fp16")]; + fp16 var_2488_promoted_2_to_fp16 = const()[name = string("op_2488_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_3186_cast_fp16 = pow(x = x_121_cast_fp16, y = var_2488_promoted_2_to_fp16)[name = string("op_3186_cast_fp16")]; + tensor var_3188_axes_0 = const()[name = string("op_3188_axes_0"), val = tensor([-1])]; + bool var_3188_keep_dims_0 = const()[name = string("op_3188_keep_dims_0"), val = bool(true)]; + tensor var_3188_cast_fp16 = reduce_mean(axes = var_3188_axes_0, keep_dims = var_3188_keep_dims_0, x = var_3186_cast_fp16)[name = string("op_3188_cast_fp16")]; + fp16 var_3189_to_fp16 = const()[name = string("op_3189_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3190_cast_fp16 = add(x = var_3188_cast_fp16, y = var_3189_to_fp16)[name = string("op_3190_cast_fp16")]; + fp32 var_3191_epsilon_0 = const()[name = string("op_3191_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3191_cast_fp16 = rsqrt(epsilon = var_3191_epsilon_0, x = var_3190_cast_fp16)[name = string("op_3191_cast_fp16")]; + tensor x_123_cast_fp16 = mul(x = x_121_cast_fp16, y = var_3191_cast_fp16)[name = string("x_123_cast_fp16")]; + tensor var_3194_to_fp16 = const()[name = string("op_3194_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(982689856)))]; + tensor input_67_cast_fp16 = mul(x = x_123_cast_fp16, y = var_3194_to_fp16)[name = string("input_67_cast_fp16")]; + tensor layers_4_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(982695040)))]; + tensor linear_36_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_4_mlp_gate_proj_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_36_cast_fp16")]; + tensor var_3201_cast_fp16 = silu(x = linear_36_cast_fp16)[name = string("op_3201_cast_fp16")]; + tensor layers_4_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1029881024)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_4_mlp_up_proj_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_37_cast_fp16")]; + tensor input_71_cast_fp16 = mul(x = var_3201_cast_fp16, y = linear_37_cast_fp16)[name = string("input_71_cast_fp16")]; + tensor layers_4_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1077067008)))]; + tensor linear_38_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_4_mlp_down_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_38_cast_fp16")]; + tensor x_125_cast_fp16 = add(x = x_121_cast_fp16, y = linear_38_cast_fp16)[name = string("x_125_cast_fp16")]; + int32 var_3221 = const()[name = string("op_3221"), val = int32(-2)]; + int32 var_3242 = const()[name = string("op_3242"), val = int32(-1)]; + fp16 var_3241_promoted_to_fp16 = const()[name = string("op_3241_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_3251_cast_fp16 = pow(x = x_125_cast_fp16, y = var_3241_promoted_to_fp16)[name = string("op_3251_cast_fp16")]; + tensor var_3253_axes_0 = const()[name = string("op_3253_axes_0"), val = tensor([-1])]; + bool var_3253_keep_dims_0 = const()[name = string("op_3253_keep_dims_0"), val = bool(true)]; + tensor var_3253_cast_fp16 = reduce_mean(axes = var_3253_axes_0, keep_dims = var_3253_keep_dims_0, x = var_3251_cast_fp16)[name = string("op_3253_cast_fp16")]; + fp16 var_3254_to_fp16 = const()[name = string("op_3254_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3255_cast_fp16 = add(x = var_3253_cast_fp16, y = var_3254_to_fp16)[name = string("op_3255_cast_fp16")]; + fp32 var_3256_epsilon_0 = const()[name = string("op_3256_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3256_cast_fp16 = rsqrt(epsilon = var_3256_epsilon_0, x = var_3255_cast_fp16)[name = string("op_3256_cast_fp16")]; + tensor x_127_cast_fp16 = mul(x = x_125_cast_fp16, y = var_3256_cast_fp16)[name = string("x_127_cast_fp16")]; + tensor var_3259_to_fp16 = const()[name = string("op_3259_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1124252992)))]; + tensor x_129_cast_fp16 = mul(x = x_127_cast_fp16, y = var_3259_to_fp16)[name = string("x_129_cast_fp16")]; + tensor layers_5_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_5_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1124258176)))]; + tensor linear_39_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_linear_attn_in_proj_qkv_weight_to_fp16, x = x_129_cast_fp16)[name = string("linear_39_cast_fp16")]; + tensor input_73_perm_0 = const()[name = string("input_73_perm_0"), val = tensor([0, 2, 1])]; + tensor input_75_pad_0 = const()[name = string("input_75_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_75_mode_0 = const()[name = string("input_75_mode_0"), val = string("constant")]; + fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)]; + tensor input_73_cast_fp16 = transpose(perm = input_73_perm_0, x = linear_39_cast_fp16)[name = string("transpose_19")]; + tensor input_75_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = input_75_mode_0, pad = input_75_pad_0, x = input_73_cast_fp16)[name = string("input_75_cast_fp16")]; + string input_77_pad_type_0 = const()[name = string("input_77_pad_type_0"), val = string("valid")]; + int32 input_77_groups_0 = const()[name = string("input_77_groups_0"), val = int32(8192)]; + tensor input_77_strides_0 = const()[name = string("input_77_strides_0"), val = tensor([1])]; + tensor input_77_pad_0 = const()[name = string("input_77_pad_0"), val = tensor([0, 0])]; + tensor input_77_dilations_0 = const()[name = string("input_77_dilations_0"), val = tensor([1])]; + tensor layers_5_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_5_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1166201280)))]; + tensor input_77_cast_fp16 = conv(dilations = input_77_dilations_0, groups = input_77_groups_0, pad = input_77_pad_0, pad_type = input_77_pad_type_0, strides = input_77_strides_0, weight = layers_5_linear_attn_conv1d_weight_to_fp16, x = input_75_cast_fp16)[name = string("input_77_cast_fp16")]; + tensor var_3288_cast_fp16 = silu(x = input_77_cast_fp16)[name = string("op_3288_cast_fp16")]; + tensor qkv_9_perm_0 = const()[name = string("qkv_9_perm_0"), val = tensor([0, 2, 1])]; + tensor var_3290_begin_0 = const()[name = string("op_3290_begin_0"), val = tensor([0, 0, 0])]; + tensor var_3290_end_0 = const()[name = string("op_3290_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_3290_end_mask_0 = const()[name = string("op_3290_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_9_cast_fp16 = transpose(perm = qkv_9_perm_0, x = var_3288_cast_fp16)[name = string("transpose_18")]; + tensor var_3290_cast_fp16 = slice_by_index(begin = var_3290_begin_0, end = var_3290_end_0, end_mask = var_3290_end_mask_0, x = qkv_9_cast_fp16)[name = string("op_3290_cast_fp16")]; + tensor var_3291 = const()[name = string("op_3291"), val = tensor([1, 1024, 16, 128])]; + tensor q_45_cast_fp16 = reshape(shape = var_3291, x = var_3290_cast_fp16)[name = string("q_45_cast_fp16")]; + tensor var_3293_begin_0 = const()[name = string("op_3293_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_3293_end_0 = const()[name = string("op_3293_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_3293_end_mask_0 = const()[name = string("op_3293_end_mask_0"), val = tensor([true, true, false])]; + tensor var_3293_cast_fp16 = slice_by_index(begin = var_3293_begin_0, end = var_3293_end_0, end_mask = var_3293_end_mask_0, x = qkv_9_cast_fp16)[name = string("op_3293_cast_fp16")]; + tensor var_3294 = const()[name = string("op_3294"), val = tensor([1, 1024, 16, 128])]; + tensor k_39_cast_fp16 = reshape(shape = var_3294, x = var_3293_cast_fp16)[name = string("k_39_cast_fp16")]; + tensor var_3296_begin_0 = const()[name = string("op_3296_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_3296_end_0 = const()[name = string("op_3296_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_3296_end_mask_0 = const()[name = string("op_3296_end_mask_0"), val = tensor([true, true, true])]; + tensor var_3296_cast_fp16 = slice_by_index(begin = var_3296_begin_0, end = var_3296_end_0, end_mask = var_3296_end_mask_0, x = qkv_9_cast_fp16)[name = string("op_3296_cast_fp16")]; + tensor var_3297 = const()[name = string("op_3297"), val = tensor([1, 1024, 32, 128])]; + tensor v_21_cast_fp16 = reshape(shape = var_3297, x = var_3296_cast_fp16)[name = string("v_21_cast_fp16")]; + tensor layers_5_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_5_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1166266880)))]; + tensor linear_40_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_5_linear_attn_in_proj_z_weight_to_fp16, x = x_129_cast_fp16)[name = string("linear_40_cast_fp16")]; + tensor var_3301 = const()[name = string("op_3301"), val = tensor([1, 1024, 32, 128])]; + tensor z_9_cast_fp16 = reshape(shape = var_3301, x = linear_40_cast_fp16)[name = string("z_9_cast_fp16")]; + tensor layers_5_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_5_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187238464)))]; + tensor linear_41_bias_0_to_fp16 = const()[name = string("linear_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187402368)))]; + tensor linear_41_cast_fp16 = linear(bias = linear_41_bias_0_to_fp16, weight = layers_5_linear_attn_in_proj_b_weight_to_fp16, x = x_129_cast_fp16)[name = string("linear_41_cast_fp16")]; + tensor beta_17_cast_fp16 = sigmoid(x = linear_41_cast_fp16)[name = string("beta_17_cast_fp16")]; + tensor var_3310_weight_0_to_fp16 = const()[name = string("op_3310_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187402496)))]; + tensor var_3310_bias_0_to_fp16 = const()[name = string("op_3310_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187566400)))]; + tensor var_3310_cast_fp16 = linear(bias = var_3310_bias_0_to_fp16, weight = var_3310_weight_0_to_fp16, x = x_129_cast_fp16)[name = string("op_3310_cast_fp16")]; + tensor var_3311_cast_fp16 = softplus(x = var_3310_cast_fp16)[name = string("op_3311_cast_fp16")]; + tensor var_3307_to_fp16 = const()[name = string("op_3307_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187566528)))]; + tensor g_17_cast_fp16 = mul(x = var_3307_to_fp16, y = var_3311_cast_fp16)[name = string("g_17_cast_fp16")]; + tensor var_3313_cast_fp16 = mul(x = q_45_cast_fp16, y = q_45_cast_fp16)[name = string("op_3313_cast_fp16")]; + tensor var_3315_axes_0 = const()[name = string("op_3315_axes_0"), val = tensor([-1])]; + bool var_3315_keep_dims_0 = const()[name = string("op_3315_keep_dims_0"), val = bool(true)]; + tensor var_3315_cast_fp16 = reduce_sum(axes = var_3315_axes_0, keep_dims = var_3315_keep_dims_0, x = var_3313_cast_fp16)[name = string("op_3315_cast_fp16")]; + fp16 var_3316_to_fp16 = const()[name = string("op_3316_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3317_cast_fp16 = add(x = var_3315_cast_fp16, y = var_3316_to_fp16)[name = string("op_3317_cast_fp16")]; + fp32 var_3318_epsilon_0 = const()[name = string("op_3318_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3318_cast_fp16 = rsqrt(epsilon = var_3318_epsilon_0, x = var_3317_cast_fp16)[name = string("op_3318_cast_fp16")]; + tensor q_47_cast_fp16 = mul(x = q_45_cast_fp16, y = var_3318_cast_fp16)[name = string("q_47_cast_fp16")]; + tensor var_3320_cast_fp16 = mul(x = k_39_cast_fp16, y = k_39_cast_fp16)[name = string("op_3320_cast_fp16")]; + tensor var_3322_axes_0 = const()[name = string("op_3322_axes_0"), val = tensor([-1])]; + bool var_3322_keep_dims_0 = const()[name = string("op_3322_keep_dims_0"), val = bool(true)]; + tensor var_3322_cast_fp16 = reduce_sum(axes = var_3322_axes_0, keep_dims = var_3322_keep_dims_0, x = var_3320_cast_fp16)[name = string("op_3322_cast_fp16")]; + fp16 var_3323_to_fp16 = const()[name = string("op_3323_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3324_cast_fp16 = add(x = var_3322_cast_fp16, y = var_3323_to_fp16)[name = string("op_3324_cast_fp16")]; + fp32 var_3325_epsilon_0 = const()[name = string("op_3325_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3325_cast_fp16 = rsqrt(epsilon = var_3325_epsilon_0, x = var_3324_cast_fp16)[name = string("op_3325_cast_fp16")]; + tensor k_41_cast_fp16 = mul(x = k_39_cast_fp16, y = var_3325_cast_fp16)[name = string("k_41_cast_fp16")]; + tensor var_3330_axes_0 = const()[name = string("op_3330_axes_0"), val = tensor([3])]; + tensor var_3330_cast_fp16 = expand_dims(axes = var_3330_axes_0, x = q_47_cast_fp16)[name = string("op_3330_cast_fp16")]; + tensor var_3332_reps_0 = const()[name = string("op_3332_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_3332_cast_fp16 = tile(reps = var_3332_reps_0, x = var_3330_cast_fp16)[name = string("op_3332_cast_fp16")]; + tensor var_3333 = const()[name = string("op_3333"), val = tensor([1, 1024, 32, 128])]; + tensor q_49_cast_fp16 = reshape(shape = var_3333, x = var_3332_cast_fp16)[name = string("q_49_cast_fp16")]; + tensor var_3338_axes_0 = const()[name = string("op_3338_axes_0"), val = tensor([3])]; + tensor var_3338_cast_fp16 = expand_dims(axes = var_3338_axes_0, x = k_41_cast_fp16)[name = string("op_3338_cast_fp16")]; + tensor var_3340_reps_0 = const()[name = string("op_3340_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_3340_cast_fp16 = tile(reps = var_3340_reps_0, x = var_3338_cast_fp16)[name = string("op_3340_cast_fp16")]; + tensor var_3341 = const()[name = string("op_3341"), val = tensor([1, 1024, 32, 128])]; + tensor k_43_cast_fp16 = reshape(shape = var_3341, x = var_3340_cast_fp16)[name = string("k_43_cast_fp16")]; + fp16 var_3343_to_fp16 = const()[name = string("op_3343_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_51_cast_fp16 = mul(x = q_49_cast_fp16, y = var_3343_to_fp16)[name = string("q_51_cast_fp16")]; + tensor var_3345_begin_0 = const()[name = string("op_3345_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3345_end_0 = const()[name = string("op_3345_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_3345_end_mask_0 = const()[name = string("op_3345_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_3345_squeeze_mask_0 = const()[name = string("op_3345_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_3345_cast_fp16 = slice_by_index(begin = var_3345_begin_0, end = var_3345_end_0, end_mask = var_3345_end_mask_0, squeeze_mask = var_3345_squeeze_mask_0, x = q_51_cast_fp16)[name = string("op_3345_cast_fp16")]; + tensor var_3346 = const()[name = string("op_3346"), val = tensor([1, 0, 2])]; + tensor var_3348 = const()[name = string("op_3348"), val = tensor([32, 16, 64, 128])]; + tensor var_3347_cast_fp16 = transpose(perm = var_3346, x = var_3345_cast_fp16)[name = string("transpose_17")]; + tensor q_53_cast_fp16 = reshape(shape = var_3348, x = var_3347_cast_fp16)[name = string("q_53_cast_fp16")]; + tensor var_3350_begin_0 = const()[name = string("op_3350_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3350_end_0 = const()[name = string("op_3350_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_3350_end_mask_0 = const()[name = string("op_3350_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_3350_squeeze_mask_0 = const()[name = string("op_3350_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_3350_cast_fp16 = slice_by_index(begin = var_3350_begin_0, end = var_3350_end_0, end_mask = var_3350_end_mask_0, squeeze_mask = var_3350_squeeze_mask_0, x = k_43_cast_fp16)[name = string("op_3350_cast_fp16")]; + tensor var_3351 = const()[name = string("op_3351"), val = tensor([1, 0, 2])]; + tensor var_3353 = const()[name = string("op_3353"), val = tensor([32, 16, 64, 128])]; + tensor var_3352_cast_fp16 = transpose(perm = var_3351, x = var_3350_cast_fp16)[name = string("transpose_16")]; + tensor k_45_cast_fp16 = reshape(shape = var_3353, x = var_3352_cast_fp16)[name = string("k_45_cast_fp16")]; + tensor var_3355_begin_0 = const()[name = string("op_3355_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3355_end_0 = const()[name = string("op_3355_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_3355_end_mask_0 = const()[name = string("op_3355_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_3355_squeeze_mask_0 = const()[name = string("op_3355_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_3355_cast_fp16 = slice_by_index(begin = var_3355_begin_0, end = var_3355_end_0, end_mask = var_3355_end_mask_0, squeeze_mask = var_3355_squeeze_mask_0, x = v_21_cast_fp16)[name = string("op_3355_cast_fp16")]; + tensor var_3356 = const()[name = string("op_3356"), val = tensor([1, 0, 2])]; + tensor var_3358 = const()[name = string("op_3358"), val = tensor([32, 16, 64, 128])]; + tensor var_3357_cast_fp16 = transpose(perm = var_3356, x = var_3355_cast_fp16)[name = string("transpose_15")]; + tensor v_23_cast_fp16 = reshape(shape = var_3358, x = var_3357_cast_fp16)[name = string("v_23_cast_fp16")]; + tensor var_3360_begin_0 = const()[name = string("op_3360_begin_0"), val = tensor([0, 0, 0])]; + tensor var_3360_end_0 = const()[name = string("op_3360_end_0"), val = tensor([1, 1024, 32])]; + tensor var_3360_end_mask_0 = const()[name = string("op_3360_end_mask_0"), val = tensor([false, true, true])]; + tensor var_3360_squeeze_mask_0 = const()[name = string("op_3360_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_3360_cast_fp16 = slice_by_index(begin = var_3360_begin_0, end = var_3360_end_0, end_mask = var_3360_end_mask_0, squeeze_mask = var_3360_squeeze_mask_0, x = beta_17_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3361_perm_0 = const()[name = string("op_3361_perm_0"), val = tensor([1, 0])]; + tensor var_3362 = const()[name = string("op_3362"), val = tensor([32, 16, 64, 1])]; + tensor var_3361_cast_fp16 = transpose(perm = var_3361_perm_0, x = var_3360_cast_fp16)[name = string("transpose_14")]; + tensor beta_19_cast_fp16 = reshape(shape = var_3362, x = var_3361_cast_fp16)[name = string("beta_19_cast_fp16")]; + tensor var_3364_begin_0 = const()[name = string("op_3364_begin_0"), val = tensor([0, 0, 0])]; + tensor var_3364_end_0 = const()[name = string("op_3364_end_0"), val = tensor([1, 1024, 32])]; + tensor var_3364_end_mask_0 = const()[name = string("op_3364_end_mask_0"), val = tensor([false, true, true])]; + tensor var_3364_squeeze_mask_0 = const()[name = string("op_3364_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_3364_cast_fp16 = slice_by_index(begin = var_3364_begin_0, end = var_3364_end_0, end_mask = var_3364_end_mask_0, squeeze_mask = var_3364_squeeze_mask_0, x = g_17_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3365_perm_0 = const()[name = string("op_3365_perm_0"), val = tensor([1, 0])]; + tensor var_3366 = const()[name = string("op_3366"), val = tensor([32, 16, 64])]; + tensor var_3365_cast_fp16 = transpose(perm = var_3365_perm_0, x = var_3364_cast_fp16)[name = string("transpose_13")]; + tensor g_19_cast_fp16 = reshape(shape = var_3366, x = var_3365_cast_fp16)[name = string("g_19_cast_fp16")]; + tensor g_col_9_axes_0 = const()[name = string("g_col_9_axes_0"), val = tensor([-1])]; + tensor g_col_9_cast_fp16 = expand_dims(axes = g_col_9_axes_0, x = g_19_cast_fp16)[name = string("g_col_9_cast_fp16")]; + bool var_3369_transpose_x_0 = const()[name = string("op_3369_transpose_x_0"), val = bool(false)]; + bool var_3369_transpose_y_0 = const()[name = string("op_3369_transpose_y_0"), val = bool(false)]; + tensor var_3369_cast_fp16 = matmul(transpose_x = var_3369_transpose_x_0, transpose_y = var_3369_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_9_cast_fp16)[name = string("op_3369_cast_fp16")]; + tensor cum_9_axes_0 = const()[name = string("cum_9_axes_0"), val = tensor([-1])]; + tensor cum_9_cast_fp16 = squeeze(axes = cum_9_axes_0, x = var_3369_cast_fp16)[name = string("cum_9_cast_fp16")]; + bool var_3371_transpose_x_0 = const()[name = string("op_3371_transpose_x_0"), val = bool(false)]; + bool var_3371_transpose_y_0 = const()[name = string("op_3371_transpose_y_0"), val = bool(false)]; + tensor var_3371_cast_fp16 = matmul(transpose_x = var_3371_transpose_x_0, transpose_y = var_3371_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_9_cast_fp16)[name = string("op_3371_cast_fp16")]; + tensor to_end_9_axes_0 = const()[name = string("to_end_9_axes_0"), val = tensor([-1])]; + tensor to_end_9_cast_fp16 = squeeze(axes = to_end_9_axes_0, x = var_3371_cast_fp16)[name = string("to_end_9_cast_fp16")]; + bool var_3373_transpose_x_0 = const()[name = string("op_3373_transpose_x_0"), val = bool(false)]; + bool var_3373_transpose_y_0 = const()[name = string("op_3373_transpose_y_0"), val = bool(false)]; + tensor var_3373_cast_fp16 = matmul(transpose_x = var_3373_transpose_x_0, transpose_y = var_3373_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_9_cast_fp16)[name = string("op_3373_cast_fp16")]; + tensor var_3374 = const()[name = string("op_3374"), val = tensor([32, 16, 64, 64])]; + tensor pair_57_cast_fp16 = reshape(shape = var_3374, x = var_3373_cast_fp16)[name = string("pair_57_cast_fp16")]; + tensor var_3376_cast_fp16 = exp(x = pair_57_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor pair_decay_9_cast_fp16 = mul(x = var_3376_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_9_cast_fp16")]; + tensor k_beta_9_cast_fp16 = mul(x = k_45_cast_fp16, y = beta_19_cast_fp16)[name = string("k_beta_9_cast_fp16")]; + tensor v_beta_9_cast_fp16 = mul(x = v_23_cast_fp16, y = beta_19_cast_fp16)[name = string("v_beta_9_cast_fp16")]; + bool var_3381_transpose_x_1 = const()[name = string("op_3381_transpose_x_1"), val = bool(false)]; + bool var_3381_transpose_y_1 = const()[name = string("op_3381_transpose_y_1"), val = bool(true)]; + tensor var_3381_cast_fp16 = matmul(transpose_x = var_3381_transpose_x_1, transpose_y = var_3381_transpose_y_1, x = k_beta_9_cast_fp16, y = k_45_cast_fp16)[name = string("op_3381_cast_fp16")]; + tensor kkt_9_cast_fp16 = mul(x = var_3381_cast_fp16, y = pair_decay_9_cast_fp16)[name = string("kkt_9_cast_fp16")]; + bool var_3384_transpose_x_1 = const()[name = string("op_3384_transpose_x_1"), val = bool(false)]; + bool var_3384_transpose_y_1 = const()[name = string("op_3384_transpose_y_1"), val = bool(true)]; + tensor var_3384_cast_fp16 = matmul(transpose_x = var_3384_transpose_x_1, transpose_y = var_3384_transpose_y_1, x = q_53_cast_fp16, y = k_45_cast_fp16)[name = string("op_3384_cast_fp16")]; + tensor attn_intra_9_cast_fp16 = mul(x = var_3384_cast_fp16, y = pair_decay_9_cast_fp16)[name = string("attn_intra_9_cast_fp16")]; + tensor var_3386_cast_fp16 = mul(x = kkt_9_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_3386_cast_fp16")]; + tensor t_17_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_3386_cast_fp16)[name = string("t_17_cast_fp16")]; + tensor var_3390 = const()[name = string("op_3390"), val = tensor([-1, 64, 64])]; + tensor t_19_cast_fp16 = reshape(shape = var_3390, x = t_17_cast_fp16)[name = string("t_19_cast_fp16")]; + tensor var_3395 = const()[name = string("op_3395"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_3396_cast_fp16 = reshape(shape = var_3395, x = t_19_cast_fp16)[name = string("op_3396_cast_fp16")]; + tensor var_3401_cast_fp16 = mul(x = var_3396_cast_fp16, y = var_216_to_fp16)[name = string("op_3401_cast_fp16")]; + tensor blocks_49_axes_0 = const()[name = string("blocks_49_axes_0"), val = tensor([-2])]; + bool blocks_49_keep_dims_0 = const()[name = string("blocks_49_keep_dims_0"), val = bool(false)]; + tensor blocks_49_cast_fp16 = reduce_sum(axes = blocks_49_axes_0, keep_dims = blocks_49_keep_dims_0, x = var_3401_cast_fp16)[name = string("blocks_49_cast_fp16")]; + tensor var_3404_begin_0 = const()[name = string("op_3404_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3404_end_0 = const()[name = string("op_3404_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_3404_end_mask_0 = const()[name = string("op_3404_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3404_cast_fp16 = slice_by_index(begin = var_3404_begin_0, end = var_3404_end_0, end_mask = var_3404_end_mask_0, x = blocks_49_cast_fp16)[name = string("op_3404_cast_fp16")]; + tensor x_131_begin_0 = const()[name = string("x_131_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_131_end_0 = const()[name = string("x_131_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_131_end_mask_0 = const()[name = string("x_131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_131_cast_fp16 = slice_by_index(begin = x_131_begin_0, end = x_131_end_0, end_mask = x_131_end_mask_0, x = var_3404_cast_fp16)[name = string("x_131_cast_fp16")]; + bool var_3414_transpose_x_0 = const()[name = string("op_3414_transpose_x_0"), val = bool(false)]; + bool var_3414_transpose_y_0 = const()[name = string("op_3414_transpose_y_0"), val = bool(false)]; + tensor var_3414_cast_fp16 = matmul(transpose_x = var_3414_transpose_x_0, transpose_y = var_3414_transpose_y_0, x = x_131_cast_fp16, y = var_224_to_fp16)[name = string("op_3414_cast_fp16")]; + bool var_3415_transpose_x_0 = const()[name = string("op_3415_transpose_x_0"), val = bool(false)]; + bool var_3415_transpose_y_0 = const()[name = string("op_3415_transpose_y_0"), val = bool(false)]; + tensor var_3415_cast_fp16 = matmul(transpose_x = var_3415_transpose_x_0, transpose_y = var_3415_transpose_y_0, x = var_224_to_fp16, y = var_3414_cast_fp16)[name = string("op_3415_cast_fp16")]; + fp16 const_59_promoted_to_fp16 = const()[name = string("const_59_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_49_cast_fp16 = mul(x = var_3415_cast_fp16, y = const_59_promoted_to_fp16)[name = string("lower_49_cast_fp16")]; + bool bottom_49_interleave_0 = const()[name = string("bottom_49_interleave_0"), val = bool(false)]; + tensor bottom_49_cast_fp16 = concat(axis = var_3242, interleave = bottom_49_interleave_0, values = (lower_49_cast_fp16, var_224_to_fp16))[name = string("bottom_49_cast_fp16")]; + bool inv_67_interleave_0 = const()[name = string("inv_67_interleave_0"), val = bool(false)]; + tensor inv_67_cast_fp16 = concat(axis = var_3221, interleave = inv_67_interleave_0, values = (top_1_to_fp16, bottom_49_cast_fp16))[name = string("inv_67_cast_fp16")]; + tensor var_3424 = const()[name = string("op_3424"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_3425_cast_fp16 = reshape(shape = var_3424, x = t_19_cast_fp16)[name = string("op_3425_cast_fp16")]; + tensor var_3430_cast_fp16 = mul(x = var_3425_cast_fp16, y = var_245_to_fp16)[name = string("op_3430_cast_fp16")]; + tensor blocks_51_axes_0 = const()[name = string("blocks_51_axes_0"), val = tensor([-2])]; + bool blocks_51_keep_dims_0 = const()[name = string("blocks_51_keep_dims_0"), val = bool(false)]; + tensor blocks_51_cast_fp16 = reduce_sum(axes = blocks_51_axes_0, keep_dims = blocks_51_keep_dims_0, x = var_3430_cast_fp16)[name = string("blocks_51_cast_fp16")]; + tensor var_3433_begin_0 = const()[name = string("op_3433_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_3433_end_0 = const()[name = string("op_3433_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_3433_end_mask_0 = const()[name = string("op_3433_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3433_cast_fp16 = slice_by_index(begin = var_3433_begin_0, end = var_3433_end_0, end_mask = var_3433_end_mask_0, x = blocks_51_cast_fp16)[name = string("op_3433_cast_fp16")]; + tensor x_133_begin_0 = const()[name = string("x_133_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_133_end_0 = const()[name = string("x_133_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_133_end_mask_0 = const()[name = string("x_133_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_133_cast_fp16 = slice_by_index(begin = x_133_begin_0, end = x_133_end_0, end_mask = x_133_end_mask_0, x = var_3433_cast_fp16)[name = string("x_133_cast_fp16")]; + tensor var_3435 = const()[name = string("op_3435"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_61_cast_fp16 = reshape(shape = var_3435, x = inv_67_cast_fp16)[name = string("pair_61_cast_fp16")]; + tensor var_3437_begin_0 = const()[name = string("op_3437_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_3437_end_0 = const()[name = string("op_3437_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_3437_end_mask_0 = const()[name = string("op_3437_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3437_squeeze_mask_0 = const()[name = string("op_3437_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3437_cast_fp16 = slice_by_index(begin = var_3437_begin_0, end = var_3437_end_0, end_mask = var_3437_end_mask_0, squeeze_mask = var_3437_squeeze_mask_0, x = pair_61_cast_fp16)[name = string("op_3437_cast_fp16")]; + tensor var_3440_begin_0 = const()[name = string("op_3440_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_3440_end_0 = const()[name = string("op_3440_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_3440_end_mask_0 = const()[name = string("op_3440_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3440_squeeze_mask_0 = const()[name = string("op_3440_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3440_cast_fp16 = slice_by_index(begin = var_3440_begin_0, end = var_3440_end_0, end_mask = var_3440_end_mask_0, squeeze_mask = var_3440_squeeze_mask_0, x = pair_61_cast_fp16)[name = string("op_3440_cast_fp16")]; + bool var_3443_transpose_x_0 = const()[name = string("op_3443_transpose_x_0"), val = bool(false)]; + bool var_3443_transpose_y_0 = const()[name = string("op_3443_transpose_y_0"), val = bool(false)]; + tensor var_3443_cast_fp16 = matmul(transpose_x = var_3443_transpose_x_0, transpose_y = var_3443_transpose_y_0, x = x_133_cast_fp16, y = var_3437_cast_fp16)[name = string("op_3443_cast_fp16")]; + bool var_3444_transpose_x_0 = const()[name = string("op_3444_transpose_x_0"), val = bool(false)]; + bool var_3444_transpose_y_0 = const()[name = string("op_3444_transpose_y_0"), val = bool(false)]; + tensor var_3444_cast_fp16 = matmul(transpose_x = var_3444_transpose_x_0, transpose_y = var_3444_transpose_y_0, x = var_3440_cast_fp16, y = var_3443_cast_fp16)[name = string("op_3444_cast_fp16")]; + fp16 const_60_promoted_to_fp16 = const()[name = string("const_60_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_51_cast_fp16 = mul(x = var_3444_cast_fp16, y = const_60_promoted_to_fp16)[name = string("lower_51_cast_fp16")]; + bool top_51_interleave_0 = const()[name = string("top_51_interleave_0"), val = bool(false)]; + tensor top_51_cast_fp16 = concat(axis = var_3242, interleave = top_51_interleave_0, values = (var_3437_cast_fp16, var_262_to_fp16))[name = string("top_51_cast_fp16")]; + bool bottom_51_interleave_0 = const()[name = string("bottom_51_interleave_0"), val = bool(false)]; + tensor bottom_51_cast_fp16 = concat(axis = var_3242, interleave = bottom_51_interleave_0, values = (lower_51_cast_fp16, var_3440_cast_fp16))[name = string("bottom_51_cast_fp16")]; + bool inv_69_interleave_0 = const()[name = string("inv_69_interleave_0"), val = bool(false)]; + tensor inv_69_cast_fp16 = concat(axis = var_3221, interleave = inv_69_interleave_0, values = (top_51_cast_fp16, bottom_51_cast_fp16))[name = string("inv_69_cast_fp16")]; + tensor var_3453 = const()[name = string("op_3453"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_3454_cast_fp16 = reshape(shape = var_3453, x = t_19_cast_fp16)[name = string("op_3454_cast_fp16")]; + tensor var_3458_to_fp16 = const()[name = string("op_3458_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187566656)))]; + tensor var_3459_cast_fp16 = mul(x = var_3454_cast_fp16, y = var_3458_to_fp16)[name = string("op_3459_cast_fp16")]; + tensor blocks_53_axes_0 = const()[name = string("blocks_53_axes_0"), val = tensor([-2])]; + bool blocks_53_keep_dims_0 = const()[name = string("blocks_53_keep_dims_0"), val = bool(false)]; + tensor blocks_53_cast_fp16 = reduce_sum(axes = blocks_53_axes_0, keep_dims = blocks_53_keep_dims_0, x = var_3459_cast_fp16)[name = string("blocks_53_cast_fp16")]; + tensor var_3462_begin_0 = const()[name = string("op_3462_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_3462_end_0 = const()[name = string("op_3462_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_3462_end_mask_0 = const()[name = string("op_3462_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3462_cast_fp16 = slice_by_index(begin = var_3462_begin_0, end = var_3462_end_0, end_mask = var_3462_end_mask_0, x = blocks_53_cast_fp16)[name = string("op_3462_cast_fp16")]; + tensor x_135_begin_0 = const()[name = string("x_135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_135_end_0 = const()[name = string("x_135_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_135_end_mask_0 = const()[name = string("x_135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_135_cast_fp16 = slice_by_index(begin = x_135_begin_0, end = x_135_end_0, end_mask = x_135_end_mask_0, x = var_3462_cast_fp16)[name = string("x_135_cast_fp16")]; + tensor var_3464 = const()[name = string("op_3464"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_63_cast_fp16 = reshape(shape = var_3464, x = inv_69_cast_fp16)[name = string("pair_63_cast_fp16")]; + tensor var_3466_begin_0 = const()[name = string("op_3466_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_3466_end_0 = const()[name = string("op_3466_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_3466_end_mask_0 = const()[name = string("op_3466_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3466_squeeze_mask_0 = const()[name = string("op_3466_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3466_cast_fp16 = slice_by_index(begin = var_3466_begin_0, end = var_3466_end_0, end_mask = var_3466_end_mask_0, squeeze_mask = var_3466_squeeze_mask_0, x = pair_63_cast_fp16)[name = string("op_3466_cast_fp16")]; + tensor var_3469_begin_0 = const()[name = string("op_3469_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_3469_end_0 = const()[name = string("op_3469_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_3469_end_mask_0 = const()[name = string("op_3469_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3469_squeeze_mask_0 = const()[name = string("op_3469_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3469_cast_fp16 = slice_by_index(begin = var_3469_begin_0, end = var_3469_end_0, end_mask = var_3469_end_mask_0, squeeze_mask = var_3469_squeeze_mask_0, x = pair_63_cast_fp16)[name = string("op_3469_cast_fp16")]; + bool var_3472_transpose_x_0 = const()[name = string("op_3472_transpose_x_0"), val = bool(false)]; + bool var_3472_transpose_y_0 = const()[name = string("op_3472_transpose_y_0"), val = bool(false)]; + tensor var_3472_cast_fp16 = matmul(transpose_x = var_3472_transpose_x_0, transpose_y = var_3472_transpose_y_0, x = x_135_cast_fp16, y = var_3466_cast_fp16)[name = string("op_3472_cast_fp16")]; + bool var_3473_transpose_x_0 = const()[name = string("op_3473_transpose_x_0"), val = bool(false)]; + bool var_3473_transpose_y_0 = const()[name = string("op_3473_transpose_y_0"), val = bool(false)]; + tensor var_3473_cast_fp16 = matmul(transpose_x = var_3473_transpose_x_0, transpose_y = var_3473_transpose_y_0, x = var_3469_cast_fp16, y = var_3472_cast_fp16)[name = string("op_3473_cast_fp16")]; + fp16 const_61_promoted_to_fp16 = const()[name = string("const_61_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_53_cast_fp16 = mul(x = var_3473_cast_fp16, y = const_61_promoted_to_fp16)[name = string("lower_53_cast_fp16")]; + bool top_53_interleave_0 = const()[name = string("top_53_interleave_0"), val = bool(false)]; + tensor top_53_cast_fp16 = concat(axis = var_3242, interleave = top_53_interleave_0, values = (var_3466_cast_fp16, var_291_to_fp16))[name = string("top_53_cast_fp16")]; + bool bottom_53_interleave_0 = const()[name = string("bottom_53_interleave_0"), val = bool(false)]; + tensor bottom_53_cast_fp16 = concat(axis = var_3242, interleave = bottom_53_interleave_0, values = (lower_53_cast_fp16, var_3469_cast_fp16))[name = string("bottom_53_cast_fp16")]; + bool inv_71_interleave_0 = const()[name = string("inv_71_interleave_0"), val = bool(false)]; + tensor inv_71_cast_fp16 = concat(axis = var_3221, interleave = inv_71_interleave_0, values = (top_53_cast_fp16, bottom_53_cast_fp16))[name = string("inv_71_cast_fp16")]; + tensor var_3482 = const()[name = string("op_3482"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_3483_cast_fp16 = reshape(shape = var_3482, x = t_19_cast_fp16)[name = string("op_3483_cast_fp16")]; + tensor var_3487_to_fp16 = const()[name = string("op_3487_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187566848)))]; + tensor var_3488_cast_fp16 = mul(x = var_3483_cast_fp16, y = var_3487_to_fp16)[name = string("op_3488_cast_fp16")]; + tensor blocks_55_axes_0 = const()[name = string("blocks_55_axes_0"), val = tensor([-2])]; + bool blocks_55_keep_dims_0 = const()[name = string("blocks_55_keep_dims_0"), val = bool(false)]; + tensor blocks_55_cast_fp16 = reduce_sum(axes = blocks_55_axes_0, keep_dims = blocks_55_keep_dims_0, x = var_3488_cast_fp16)[name = string("blocks_55_cast_fp16")]; + tensor var_3491_begin_0 = const()[name = string("op_3491_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_3491_end_0 = const()[name = string("op_3491_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_3491_end_mask_0 = const()[name = string("op_3491_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3491_cast_fp16 = slice_by_index(begin = var_3491_begin_0, end = var_3491_end_0, end_mask = var_3491_end_mask_0, x = blocks_55_cast_fp16)[name = string("op_3491_cast_fp16")]; + tensor x_137_begin_0 = const()[name = string("x_137_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_137_end_0 = const()[name = string("x_137_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_137_end_mask_0 = const()[name = string("x_137_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_137_cast_fp16 = slice_by_index(begin = x_137_begin_0, end = x_137_end_0, end_mask = x_137_end_mask_0, x = var_3491_cast_fp16)[name = string("x_137_cast_fp16")]; + tensor var_3493 = const()[name = string("op_3493"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_65_cast_fp16 = reshape(shape = var_3493, x = inv_71_cast_fp16)[name = string("pair_65_cast_fp16")]; + tensor var_3495_begin_0 = const()[name = string("op_3495_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_3495_end_0 = const()[name = string("op_3495_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_3495_end_mask_0 = const()[name = string("op_3495_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3495_squeeze_mask_0 = const()[name = string("op_3495_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3495_cast_fp16 = slice_by_index(begin = var_3495_begin_0, end = var_3495_end_0, end_mask = var_3495_end_mask_0, squeeze_mask = var_3495_squeeze_mask_0, x = pair_65_cast_fp16)[name = string("op_3495_cast_fp16")]; + tensor var_3498_begin_0 = const()[name = string("op_3498_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_3498_end_0 = const()[name = string("op_3498_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_3498_end_mask_0 = const()[name = string("op_3498_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3498_squeeze_mask_0 = const()[name = string("op_3498_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3498_cast_fp16 = slice_by_index(begin = var_3498_begin_0, end = var_3498_end_0, end_mask = var_3498_end_mask_0, squeeze_mask = var_3498_squeeze_mask_0, x = pair_65_cast_fp16)[name = string("op_3498_cast_fp16")]; + bool var_3501_transpose_x_0 = const()[name = string("op_3501_transpose_x_0"), val = bool(false)]; + bool var_3501_transpose_y_0 = const()[name = string("op_3501_transpose_y_0"), val = bool(false)]; + tensor var_3501_cast_fp16 = matmul(transpose_x = var_3501_transpose_x_0, transpose_y = var_3501_transpose_y_0, x = x_137_cast_fp16, y = var_3495_cast_fp16)[name = string("op_3501_cast_fp16")]; + bool var_3502_transpose_x_0 = const()[name = string("op_3502_transpose_x_0"), val = bool(false)]; + bool var_3502_transpose_y_0 = const()[name = string("op_3502_transpose_y_0"), val = bool(false)]; + tensor var_3502_cast_fp16 = matmul(transpose_x = var_3502_transpose_x_0, transpose_y = var_3502_transpose_y_0, x = var_3498_cast_fp16, y = var_3501_cast_fp16)[name = string("op_3502_cast_fp16")]; + fp16 const_62_promoted_to_fp16 = const()[name = string("const_62_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_55_cast_fp16 = mul(x = var_3502_cast_fp16, y = const_62_promoted_to_fp16)[name = string("lower_55_cast_fp16")]; + bool top_55_interleave_0 = const()[name = string("top_55_interleave_0"), val = bool(false)]; + tensor top_55_cast_fp16 = concat(axis = var_3242, interleave = top_55_interleave_0, values = (var_3495_cast_fp16, var_320_to_fp16))[name = string("top_55_cast_fp16")]; + bool bottom_55_interleave_0 = const()[name = string("bottom_55_interleave_0"), val = bool(false)]; + tensor bottom_55_cast_fp16 = concat(axis = var_3242, interleave = bottom_55_interleave_0, values = (lower_55_cast_fp16, var_3498_cast_fp16))[name = string("bottom_55_cast_fp16")]; + bool inv_73_interleave_0 = const()[name = string("inv_73_interleave_0"), val = bool(false)]; + tensor inv_73_cast_fp16 = concat(axis = var_3221, interleave = inv_73_interleave_0, values = (top_55_cast_fp16, bottom_55_cast_fp16))[name = string("inv_73_cast_fp16")]; + tensor var_3511 = const()[name = string("op_3511"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_3512_cast_fp16 = reshape(shape = var_3511, x = t_19_cast_fp16)[name = string("op_3512_cast_fp16")]; + tensor var_3516_to_fp16 = const()[name = string("op_3516_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_3517_cast_fp16 = mul(x = var_3512_cast_fp16, y = var_3516_to_fp16)[name = string("op_3517_cast_fp16")]; + tensor blocks_57_axes_0 = const()[name = string("blocks_57_axes_0"), val = tensor([-2])]; + bool blocks_57_keep_dims_0 = const()[name = string("blocks_57_keep_dims_0"), val = bool(false)]; + tensor blocks_57_cast_fp16 = reduce_sum(axes = blocks_57_axes_0, keep_dims = blocks_57_keep_dims_0, x = var_3517_cast_fp16)[name = string("blocks_57_cast_fp16")]; + tensor var_3520_begin_0 = const()[name = string("op_3520_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_3520_end_0 = const()[name = string("op_3520_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_3520_end_mask_0 = const()[name = string("op_3520_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3520_cast_fp16 = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = blocks_57_cast_fp16)[name = string("op_3520_cast_fp16")]; + tensor x_139_begin_0 = const()[name = string("x_139_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_139_end_0 = const()[name = string("x_139_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_139_end_mask_0 = const()[name = string("x_139_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_139_cast_fp16 = slice_by_index(begin = x_139_begin_0, end = x_139_end_0, end_mask = x_139_end_mask_0, x = var_3520_cast_fp16)[name = string("x_139_cast_fp16")]; + tensor var_3522 = const()[name = string("op_3522"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_67_cast_fp16 = reshape(shape = var_3522, x = inv_73_cast_fp16)[name = string("pair_67_cast_fp16")]; + tensor var_3524_begin_0 = const()[name = string("op_3524_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_3524_end_0 = const()[name = string("op_3524_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_3524_end_mask_0 = const()[name = string("op_3524_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3524_squeeze_mask_0 = const()[name = string("op_3524_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3524_cast_fp16 = slice_by_index(begin = var_3524_begin_0, end = var_3524_end_0, end_mask = var_3524_end_mask_0, squeeze_mask = var_3524_squeeze_mask_0, x = pair_67_cast_fp16)[name = string("op_3524_cast_fp16")]; + tensor var_3527_begin_0 = const()[name = string("op_3527_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_3527_end_0 = const()[name = string("op_3527_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_3527_end_mask_0 = const()[name = string("op_3527_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3527_squeeze_mask_0 = const()[name = string("op_3527_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3527_cast_fp16 = slice_by_index(begin = var_3527_begin_0, end = var_3527_end_0, end_mask = var_3527_end_mask_0, squeeze_mask = var_3527_squeeze_mask_0, x = pair_67_cast_fp16)[name = string("op_3527_cast_fp16")]; + bool var_3530_transpose_x_0 = const()[name = string("op_3530_transpose_x_0"), val = bool(false)]; + bool var_3530_transpose_y_0 = const()[name = string("op_3530_transpose_y_0"), val = bool(false)]; + tensor var_3530_cast_fp16 = matmul(transpose_x = var_3530_transpose_x_0, transpose_y = var_3530_transpose_y_0, x = x_139_cast_fp16, y = var_3524_cast_fp16)[name = string("op_3530_cast_fp16")]; + bool var_3531_transpose_x_0 = const()[name = string("op_3531_transpose_x_0"), val = bool(false)]; + bool var_3531_transpose_y_0 = const()[name = string("op_3531_transpose_y_0"), val = bool(false)]; + tensor var_3531_cast_fp16 = matmul(transpose_x = var_3531_transpose_x_0, transpose_y = var_3531_transpose_y_0, x = var_3527_cast_fp16, y = var_3530_cast_fp16)[name = string("op_3531_cast_fp16")]; + fp16 const_63_promoted_to_fp16 = const()[name = string("const_63_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_57_cast_fp16 = mul(x = var_3531_cast_fp16, y = const_63_promoted_to_fp16)[name = string("lower_57_cast_fp16")]; + bool top_57_interleave_0 = const()[name = string("top_57_interleave_0"), val = bool(false)]; + tensor top_57_cast_fp16 = concat(axis = var_3242, interleave = top_57_interleave_0, values = (var_3524_cast_fp16, var_349_to_fp16))[name = string("top_57_cast_fp16")]; + bool bottom_57_interleave_0 = const()[name = string("bottom_57_interleave_0"), val = bool(false)]; + tensor bottom_57_cast_fp16 = concat(axis = var_3242, interleave = bottom_57_interleave_0, values = (lower_57_cast_fp16, var_3527_cast_fp16))[name = string("bottom_57_cast_fp16")]; + bool inv_75_interleave_0 = const()[name = string("inv_75_interleave_0"), val = bool(false)]; + tensor inv_75_cast_fp16 = concat(axis = var_3221, interleave = inv_75_interleave_0, values = (top_57_cast_fp16, bottom_57_cast_fp16))[name = string("inv_75_cast_fp16")]; + tensor var_3540 = const()[name = string("op_3540"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_3541_cast_fp16 = reshape(shape = var_3540, x = t_19_cast_fp16)[name = string("op_3541_cast_fp16")]; + tensor blocks_59_axes_0 = const()[name = string("blocks_59_axes_0"), val = tensor([-2])]; + bool blocks_59_keep_dims_0 = const()[name = string("blocks_59_keep_dims_0"), val = bool(false)]; + tensor blocks_59_cast_fp16 = reduce_sum(axes = blocks_59_axes_0, keep_dims = blocks_59_keep_dims_0, x = var_3541_cast_fp16)[name = string("blocks_59_cast_fp16")]; + tensor var_3549_begin_0 = const()[name = string("op_3549_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_3549_end_0 = const()[name = string("op_3549_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_3549_end_mask_0 = const()[name = string("op_3549_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3549_cast_fp16 = slice_by_index(begin = var_3549_begin_0, end = var_3549_end_0, end_mask = var_3549_end_mask_0, x = blocks_59_cast_fp16)[name = string("op_3549_cast_fp16")]; + tensor x_141_begin_0 = const()[name = string("x_141_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_141_end_0 = const()[name = string("x_141_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_141_end_mask_0 = const()[name = string("x_141_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_141_cast_fp16 = slice_by_index(begin = x_141_begin_0, end = x_141_end_0, end_mask = x_141_end_mask_0, x = var_3549_cast_fp16)[name = string("x_141_cast_fp16")]; + tensor var_3551 = const()[name = string("op_3551"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_69_cast_fp16 = reshape(shape = var_3551, x = inv_75_cast_fp16)[name = string("pair_69_cast_fp16")]; + tensor var_3553_begin_0 = const()[name = string("op_3553_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_3553_end_0 = const()[name = string("op_3553_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_3553_end_mask_0 = const()[name = string("op_3553_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3553_squeeze_mask_0 = const()[name = string("op_3553_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3553_cast_fp16 = slice_by_index(begin = var_3553_begin_0, end = var_3553_end_0, end_mask = var_3553_end_mask_0, squeeze_mask = var_3553_squeeze_mask_0, x = pair_69_cast_fp16)[name = string("op_3553_cast_fp16")]; + tensor var_3556_begin_0 = const()[name = string("op_3556_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_3556_end_0 = const()[name = string("op_3556_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_3556_end_mask_0 = const()[name = string("op_3556_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_3556_squeeze_mask_0 = const()[name = string("op_3556_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_3556_cast_fp16 = slice_by_index(begin = var_3556_begin_0, end = var_3556_end_0, end_mask = var_3556_end_mask_0, squeeze_mask = var_3556_squeeze_mask_0, x = pair_69_cast_fp16)[name = string("op_3556_cast_fp16")]; + bool var_3559_transpose_x_0 = const()[name = string("op_3559_transpose_x_0"), val = bool(false)]; + bool var_3559_transpose_y_0 = const()[name = string("op_3559_transpose_y_0"), val = bool(false)]; + tensor var_3559_cast_fp16 = matmul(transpose_x = var_3559_transpose_x_0, transpose_y = var_3559_transpose_y_0, x = x_141_cast_fp16, y = var_3553_cast_fp16)[name = string("op_3559_cast_fp16")]; + bool var_3560_transpose_x_0 = const()[name = string("op_3560_transpose_x_0"), val = bool(false)]; + bool var_3560_transpose_y_0 = const()[name = string("op_3560_transpose_y_0"), val = bool(false)]; + tensor var_3560_cast_fp16 = matmul(transpose_x = var_3560_transpose_x_0, transpose_y = var_3560_transpose_y_0, x = var_3556_cast_fp16, y = var_3559_cast_fp16)[name = string("op_3560_cast_fp16")]; + fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_59_cast_fp16 = mul(x = var_3560_cast_fp16, y = const_64_promoted_to_fp16)[name = string("lower_59_cast_fp16")]; + bool top_59_interleave_0 = const()[name = string("top_59_interleave_0"), val = bool(false)]; + tensor top_59_cast_fp16 = concat(axis = var_3242, interleave = top_59_interleave_0, values = (var_3553_cast_fp16, var_378_to_fp16))[name = string("top_59_cast_fp16")]; + bool bottom_59_interleave_0 = const()[name = string("bottom_59_interleave_0"), val = bool(false)]; + tensor bottom_59_cast_fp16 = concat(axis = var_3242, interleave = bottom_59_interleave_0, values = (lower_59_cast_fp16, var_3556_cast_fp16))[name = string("bottom_59_cast_fp16")]; + bool inv_77_interleave_0 = const()[name = string("inv_77_interleave_0"), val = bool(false)]; + tensor inv_77_cast_fp16 = concat(axis = var_3221, interleave = inv_77_interleave_0, values = (top_59_cast_fp16, bottom_59_cast_fp16))[name = string("inv_77_cast_fp16")]; + tensor var_3569 = const()[name = string("op_3569"), val = tensor([32, 16, 64, 64])]; + tensor inv_79_cast_fp16 = reshape(shape = var_3569, x = inv_77_cast_fp16)[name = string("inv_79_cast_fp16")]; + bool new_v_9_transpose_x_0 = const()[name = string("new_v_9_transpose_x_0"), val = bool(false)]; + bool new_v_9_transpose_y_0 = const()[name = string("new_v_9_transpose_y_0"), val = bool(false)]; + tensor new_v_9_cast_fp16 = matmul(transpose_x = new_v_9_transpose_x_0, transpose_y = new_v_9_transpose_y_0, x = inv_79_cast_fp16, y = v_beta_9_cast_fp16)[name = string("new_v_9_cast_fp16")]; + tensor var_3572_cast_fp16 = exp(x = cum_9_cast_fp16)[name = string("op_3572_cast_fp16")]; + tensor var_3573_axes_0 = const()[name = string("op_3573_axes_0"), val = tensor([3])]; + tensor var_3573_cast_fp16 = expand_dims(axes = var_3573_axes_0, x = var_3572_cast_fp16)[name = string("op_3573_cast_fp16")]; + tensor var_3574_cast_fp16 = mul(x = k_beta_9_cast_fp16, y = var_3573_cast_fp16)[name = string("op_3574_cast_fp16")]; + bool k_cumdecay_9_transpose_x_0 = const()[name = string("k_cumdecay_9_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_9_transpose_y_0 = const()[name = string("k_cumdecay_9_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_9_cast_fp16 = matmul(transpose_x = k_cumdecay_9_transpose_x_0, transpose_y = k_cumdecay_9_transpose_y_0, x = inv_79_cast_fp16, y = var_3574_cast_fp16)[name = string("k_cumdecay_9_cast_fp16")]; + tensor q_dec_9_cast_fp16 = mul(x = q_53_cast_fp16, y = var_3573_cast_fp16)[name = string("q_dec_9_cast_fp16")]; + tensor var_3579_cast_fp16 = exp(x = to_end_9_cast_fp16)[name = string("op_3579_cast_fp16")]; + tensor var_3580_axes_0 = const()[name = string("op_3580_axes_0"), val = tensor([3])]; + tensor var_3580_cast_fp16 = expand_dims(axes = var_3580_axes_0, x = var_3579_cast_fp16)[name = string("op_3580_cast_fp16")]; + tensor k_dec_9_cast_fp16 = mul(x = k_45_cast_fp16, y = var_3580_cast_fp16)[name = string("k_dec_9_cast_fp16")]; + tensor var_3582_begin_0 = const()[name = string("op_3582_begin_0"), val = tensor([0, 0, -1])]; + tensor var_3582_end_0 = const()[name = string("op_3582_end_0"), val = tensor([32, 16, 64])]; + tensor var_3582_end_mask_0 = const()[name = string("op_3582_end_mask_0"), val = tensor([true, true, true])]; + tensor var_3582_cast_fp16 = slice_by_index(begin = var_3582_begin_0, end = var_3582_end_0, end_mask = var_3582_end_mask_0, x = cum_9_cast_fp16)[name = string("op_3582_cast_fp16")]; + tensor var_3583_cast_fp16 = exp(x = var_3582_cast_fp16)[name = string("op_3583_cast_fp16")]; + tensor chunk_decay_9_axes_0 = const()[name = string("chunk_decay_9_axes_0"), val = tensor([3])]; + tensor chunk_decay_9_cast_fp16 = expand_dims(axes = chunk_decay_9_axes_0, x = var_3583_cast_fp16)[name = string("chunk_decay_9_cast_fp16")]; + tensor var_3588_begin_0 = const()[name = string("op_3588_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3588_end_0 = const()[name = string("op_3588_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_3588_end_mask_0 = const()[name = string("op_3588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3588_squeeze_mask_0 = const()[name = string("op_3588_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3588_cast_fp16 = slice_by_index(begin = var_3588_begin_0, end = var_3588_end_0, end_mask = var_3588_end_mask_0, squeeze_mask = var_3588_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3588_cast_fp16")]; + tensor var_3590_begin_0 = const()[name = string("op_3590_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3590_end_0 = const()[name = string("op_3590_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_3590_end_mask_0 = const()[name = string("op_3590_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3590_squeeze_mask_0 = const()[name = string("op_3590_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3590_cast_fp16 = slice_by_index(begin = var_3590_begin_0, end = var_3590_end_0, end_mask = var_3590_end_mask_0, squeeze_mask = var_3590_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3590_cast_fp16")]; + bool var_3591_transpose_x_0 = const()[name = string("op_3591_transpose_x_0"), val = bool(false)]; + bool var_3591_transpose_y_0 = const()[name = string("op_3591_transpose_y_0"), val = bool(false)]; + tensor var_3591_cast_fp16 = matmul(transpose_x = var_3591_transpose_x_0, transpose_y = var_3591_transpose_y_0, x = var_3590_cast_fp16, y = state_1_to_fp16)[name = string("op_3591_cast_fp16")]; + tensor v_new_129_cast_fp16 = sub(x = var_3588_cast_fp16, y = var_3591_cast_fp16)[name = string("v_new_129_cast_fp16")]; + tensor var_3594_begin_0 = const()[name = string("op_3594_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3594_end_0 = const()[name = string("op_3594_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_3594_end_mask_0 = const()[name = string("op_3594_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3594_squeeze_mask_0 = const()[name = string("op_3594_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3594_cast_fp16 = slice_by_index(begin = var_3594_begin_0, end = var_3594_end_0, end_mask = var_3594_end_mask_0, squeeze_mask = var_3594_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3594_cast_fp16")]; + bool var_3595_transpose_x_0 = const()[name = string("op_3595_transpose_x_0"), val = bool(false)]; + bool var_3595_transpose_y_0 = const()[name = string("op_3595_transpose_y_0"), val = bool(false)]; + tensor var_3595_cast_fp16 = matmul(transpose_x = var_3595_transpose_x_0, transpose_y = var_3595_transpose_y_0, x = var_3594_cast_fp16, y = state_1_to_fp16)[name = string("op_3595_cast_fp16")]; + tensor var_3597_begin_0 = const()[name = string("op_3597_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3597_end_0 = const()[name = string("op_3597_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_3597_end_mask_0 = const()[name = string("op_3597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3597_squeeze_mask_0 = const()[name = string("op_3597_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3597_cast_fp16 = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, squeeze_mask = var_3597_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3597_cast_fp16")]; + bool var_3598_transpose_x_0 = const()[name = string("op_3598_transpose_x_0"), val = bool(false)]; + bool var_3598_transpose_y_0 = const()[name = string("op_3598_transpose_y_0"), val = bool(false)]; + tensor var_3598_cast_fp16 = matmul(transpose_x = var_3598_transpose_x_0, transpose_y = var_3598_transpose_y_0, x = var_3597_cast_fp16, y = v_new_129_cast_fp16)[name = string("op_3598_cast_fp16")]; + tensor var_3599_cast_fp16 = add(x = var_3595_cast_fp16, y = var_3598_cast_fp16)[name = string("op_3599_cast_fp16")]; + tensor var_3601_begin_0 = const()[name = string("op_3601_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3601_end_0 = const()[name = string("op_3601_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_3601_end_mask_0 = const()[name = string("op_3601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3601_squeeze_mask_0 = const()[name = string("op_3601_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3601_cast_fp16 = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, squeeze_mask = var_3601_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3601_cast_fp16")]; + tensor var_3602_cast_fp16 = mul(x = state_1_to_fp16, y = var_3601_cast_fp16)[name = string("op_3602_cast_fp16")]; + tensor var_3604_begin_0 = const()[name = string("op_3604_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3604_end_0 = const()[name = string("op_3604_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_3604_end_mask_0 = const()[name = string("op_3604_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3604_squeeze_mask_0 = const()[name = string("op_3604_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3604_cast_fp16 = slice_by_index(begin = var_3604_begin_0, end = var_3604_end_0, end_mask = var_3604_end_mask_0, squeeze_mask = var_3604_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3604_cast_fp16")]; + bool var_3606_transpose_x_1 = const()[name = string("op_3606_transpose_x_1"), val = bool(true)]; + bool var_3606_transpose_y_1 = const()[name = string("op_3606_transpose_y_1"), val = bool(false)]; + tensor var_3606_cast_fp16 = matmul(transpose_x = var_3606_transpose_x_1, transpose_y = var_3606_transpose_y_1, x = var_3604_cast_fp16, y = v_new_129_cast_fp16)[name = string("op_3606_cast_fp16")]; + tensor state_131_cast_fp16 = add(x = var_3602_cast_fp16, y = var_3606_cast_fp16)[name = string("state_131_cast_fp16")]; + tensor var_3609_begin_0 = const()[name = string("op_3609_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3609_end_0 = const()[name = string("op_3609_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_3609_end_mask_0 = const()[name = string("op_3609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3609_squeeze_mask_0 = const()[name = string("op_3609_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3609_cast_fp16 = slice_by_index(begin = var_3609_begin_0, end = var_3609_end_0, end_mask = var_3609_end_mask_0, squeeze_mask = var_3609_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3609_cast_fp16")]; + tensor var_3611_begin_0 = const()[name = string("op_3611_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3611_end_0 = const()[name = string("op_3611_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_3611_end_mask_0 = const()[name = string("op_3611_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3611_squeeze_mask_0 = const()[name = string("op_3611_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3611_cast_fp16 = slice_by_index(begin = var_3611_begin_0, end = var_3611_end_0, end_mask = var_3611_end_mask_0, squeeze_mask = var_3611_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3611_cast_fp16")]; + bool var_3612_transpose_x_0 = const()[name = string("op_3612_transpose_x_0"), val = bool(false)]; + bool var_3612_transpose_y_0 = const()[name = string("op_3612_transpose_y_0"), val = bool(false)]; + tensor var_3612_cast_fp16 = matmul(transpose_x = var_3612_transpose_x_0, transpose_y = var_3612_transpose_y_0, x = var_3611_cast_fp16, y = state_131_cast_fp16)[name = string("op_3612_cast_fp16")]; + tensor v_new_131_cast_fp16 = sub(x = var_3609_cast_fp16, y = var_3612_cast_fp16)[name = string("v_new_131_cast_fp16")]; + tensor var_3615_begin_0 = const()[name = string("op_3615_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3615_end_0 = const()[name = string("op_3615_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_3615_end_mask_0 = const()[name = string("op_3615_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3615_squeeze_mask_0 = const()[name = string("op_3615_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3615_cast_fp16 = slice_by_index(begin = var_3615_begin_0, end = var_3615_end_0, end_mask = var_3615_end_mask_0, squeeze_mask = var_3615_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3615_cast_fp16")]; + bool var_3616_transpose_x_0 = const()[name = string("op_3616_transpose_x_0"), val = bool(false)]; + bool var_3616_transpose_y_0 = const()[name = string("op_3616_transpose_y_0"), val = bool(false)]; + tensor var_3616_cast_fp16 = matmul(transpose_x = var_3616_transpose_x_0, transpose_y = var_3616_transpose_y_0, x = var_3615_cast_fp16, y = state_131_cast_fp16)[name = string("op_3616_cast_fp16")]; + tensor var_3618_begin_0 = const()[name = string("op_3618_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3618_end_0 = const()[name = string("op_3618_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_3618_end_mask_0 = const()[name = string("op_3618_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3618_squeeze_mask_0 = const()[name = string("op_3618_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3618_cast_fp16 = slice_by_index(begin = var_3618_begin_0, end = var_3618_end_0, end_mask = var_3618_end_mask_0, squeeze_mask = var_3618_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3618_cast_fp16")]; + bool var_3619_transpose_x_0 = const()[name = string("op_3619_transpose_x_0"), val = bool(false)]; + bool var_3619_transpose_y_0 = const()[name = string("op_3619_transpose_y_0"), val = bool(false)]; + tensor var_3619_cast_fp16 = matmul(transpose_x = var_3619_transpose_x_0, transpose_y = var_3619_transpose_y_0, x = var_3618_cast_fp16, y = v_new_131_cast_fp16)[name = string("op_3619_cast_fp16")]; + tensor var_3620_cast_fp16 = add(x = var_3616_cast_fp16, y = var_3619_cast_fp16)[name = string("op_3620_cast_fp16")]; + tensor var_3622_begin_0 = const()[name = string("op_3622_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3622_end_0 = const()[name = string("op_3622_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_3622_end_mask_0 = const()[name = string("op_3622_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3622_squeeze_mask_0 = const()[name = string("op_3622_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3622_cast_fp16 = slice_by_index(begin = var_3622_begin_0, end = var_3622_end_0, end_mask = var_3622_end_mask_0, squeeze_mask = var_3622_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3622_cast_fp16")]; + tensor var_3623_cast_fp16 = mul(x = state_131_cast_fp16, y = var_3622_cast_fp16)[name = string("op_3623_cast_fp16")]; + tensor var_3625_begin_0 = const()[name = string("op_3625_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_3625_end_0 = const()[name = string("op_3625_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_3625_end_mask_0 = const()[name = string("op_3625_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3625_squeeze_mask_0 = const()[name = string("op_3625_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3625_cast_fp16 = slice_by_index(begin = var_3625_begin_0, end = var_3625_end_0, end_mask = var_3625_end_mask_0, squeeze_mask = var_3625_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3625_cast_fp16")]; + bool var_3627_transpose_x_1 = const()[name = string("op_3627_transpose_x_1"), val = bool(true)]; + bool var_3627_transpose_y_1 = const()[name = string("op_3627_transpose_y_1"), val = bool(false)]; + tensor var_3627_cast_fp16 = matmul(transpose_x = var_3627_transpose_x_1, transpose_y = var_3627_transpose_y_1, x = var_3625_cast_fp16, y = v_new_131_cast_fp16)[name = string("op_3627_cast_fp16")]; + tensor state_133_cast_fp16 = add(x = var_3623_cast_fp16, y = var_3627_cast_fp16)[name = string("state_133_cast_fp16")]; + tensor var_3630_begin_0 = const()[name = string("op_3630_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3630_end_0 = const()[name = string("op_3630_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_3630_end_mask_0 = const()[name = string("op_3630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3630_squeeze_mask_0 = const()[name = string("op_3630_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3630_cast_fp16 = slice_by_index(begin = var_3630_begin_0, end = var_3630_end_0, end_mask = var_3630_end_mask_0, squeeze_mask = var_3630_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3630_cast_fp16")]; + tensor var_3632_begin_0 = const()[name = string("op_3632_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3632_end_0 = const()[name = string("op_3632_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_3632_end_mask_0 = const()[name = string("op_3632_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3632_squeeze_mask_0 = const()[name = string("op_3632_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3632_cast_fp16 = slice_by_index(begin = var_3632_begin_0, end = var_3632_end_0, end_mask = var_3632_end_mask_0, squeeze_mask = var_3632_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3632_cast_fp16")]; + bool var_3633_transpose_x_0 = const()[name = string("op_3633_transpose_x_0"), val = bool(false)]; + bool var_3633_transpose_y_0 = const()[name = string("op_3633_transpose_y_0"), val = bool(false)]; + tensor var_3633_cast_fp16 = matmul(transpose_x = var_3633_transpose_x_0, transpose_y = var_3633_transpose_y_0, x = var_3632_cast_fp16, y = state_133_cast_fp16)[name = string("op_3633_cast_fp16")]; + tensor v_new_133_cast_fp16 = sub(x = var_3630_cast_fp16, y = var_3633_cast_fp16)[name = string("v_new_133_cast_fp16")]; + tensor var_3636_begin_0 = const()[name = string("op_3636_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3636_end_0 = const()[name = string("op_3636_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_3636_end_mask_0 = const()[name = string("op_3636_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3636_squeeze_mask_0 = const()[name = string("op_3636_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3636_cast_fp16 = slice_by_index(begin = var_3636_begin_0, end = var_3636_end_0, end_mask = var_3636_end_mask_0, squeeze_mask = var_3636_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3636_cast_fp16")]; + bool var_3637_transpose_x_0 = const()[name = string("op_3637_transpose_x_0"), val = bool(false)]; + bool var_3637_transpose_y_0 = const()[name = string("op_3637_transpose_y_0"), val = bool(false)]; + tensor var_3637_cast_fp16 = matmul(transpose_x = var_3637_transpose_x_0, transpose_y = var_3637_transpose_y_0, x = var_3636_cast_fp16, y = state_133_cast_fp16)[name = string("op_3637_cast_fp16")]; + tensor var_3639_begin_0 = const()[name = string("op_3639_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3639_end_0 = const()[name = string("op_3639_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_3639_end_mask_0 = const()[name = string("op_3639_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3639_squeeze_mask_0 = const()[name = string("op_3639_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3639_cast_fp16 = slice_by_index(begin = var_3639_begin_0, end = var_3639_end_0, end_mask = var_3639_end_mask_0, squeeze_mask = var_3639_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3639_cast_fp16")]; + bool var_3640_transpose_x_0 = const()[name = string("op_3640_transpose_x_0"), val = bool(false)]; + bool var_3640_transpose_y_0 = const()[name = string("op_3640_transpose_y_0"), val = bool(false)]; + tensor var_3640_cast_fp16 = matmul(transpose_x = var_3640_transpose_x_0, transpose_y = var_3640_transpose_y_0, x = var_3639_cast_fp16, y = v_new_133_cast_fp16)[name = string("op_3640_cast_fp16")]; + tensor var_3641_cast_fp16 = add(x = var_3637_cast_fp16, y = var_3640_cast_fp16)[name = string("op_3641_cast_fp16")]; + tensor var_3643_begin_0 = const()[name = string("op_3643_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3643_end_0 = const()[name = string("op_3643_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_3643_end_mask_0 = const()[name = string("op_3643_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3643_squeeze_mask_0 = const()[name = string("op_3643_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3643_cast_fp16 = slice_by_index(begin = var_3643_begin_0, end = var_3643_end_0, end_mask = var_3643_end_mask_0, squeeze_mask = var_3643_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3643_cast_fp16")]; + tensor var_3644_cast_fp16 = mul(x = state_133_cast_fp16, y = var_3643_cast_fp16)[name = string("op_3644_cast_fp16")]; + tensor var_3646_begin_0 = const()[name = string("op_3646_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_3646_end_0 = const()[name = string("op_3646_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_3646_end_mask_0 = const()[name = string("op_3646_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3646_squeeze_mask_0 = const()[name = string("op_3646_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3646_cast_fp16 = slice_by_index(begin = var_3646_begin_0, end = var_3646_end_0, end_mask = var_3646_end_mask_0, squeeze_mask = var_3646_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3646_cast_fp16")]; + bool var_3648_transpose_x_1 = const()[name = string("op_3648_transpose_x_1"), val = bool(true)]; + bool var_3648_transpose_y_1 = const()[name = string("op_3648_transpose_y_1"), val = bool(false)]; + tensor var_3648_cast_fp16 = matmul(transpose_x = var_3648_transpose_x_1, transpose_y = var_3648_transpose_y_1, x = var_3646_cast_fp16, y = v_new_133_cast_fp16)[name = string("op_3648_cast_fp16")]; + tensor state_135_cast_fp16 = add(x = var_3644_cast_fp16, y = var_3648_cast_fp16)[name = string("state_135_cast_fp16")]; + tensor var_3651_begin_0 = const()[name = string("op_3651_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3651_end_0 = const()[name = string("op_3651_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_3651_end_mask_0 = const()[name = string("op_3651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3651_squeeze_mask_0 = const()[name = string("op_3651_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3651_cast_fp16 = slice_by_index(begin = var_3651_begin_0, end = var_3651_end_0, end_mask = var_3651_end_mask_0, squeeze_mask = var_3651_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3651_cast_fp16")]; + tensor var_3653_begin_0 = const()[name = string("op_3653_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3653_end_0 = const()[name = string("op_3653_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_3653_end_mask_0 = const()[name = string("op_3653_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3653_squeeze_mask_0 = const()[name = string("op_3653_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3653_cast_fp16 = slice_by_index(begin = var_3653_begin_0, end = var_3653_end_0, end_mask = var_3653_end_mask_0, squeeze_mask = var_3653_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3653_cast_fp16")]; + bool var_3654_transpose_x_0 = const()[name = string("op_3654_transpose_x_0"), val = bool(false)]; + bool var_3654_transpose_y_0 = const()[name = string("op_3654_transpose_y_0"), val = bool(false)]; + tensor var_3654_cast_fp16 = matmul(transpose_x = var_3654_transpose_x_0, transpose_y = var_3654_transpose_y_0, x = var_3653_cast_fp16, y = state_135_cast_fp16)[name = string("op_3654_cast_fp16")]; + tensor v_new_135_cast_fp16 = sub(x = var_3651_cast_fp16, y = var_3654_cast_fp16)[name = string("v_new_135_cast_fp16")]; + tensor var_3657_begin_0 = const()[name = string("op_3657_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3657_end_0 = const()[name = string("op_3657_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_3657_end_mask_0 = const()[name = string("op_3657_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3657_squeeze_mask_0 = const()[name = string("op_3657_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3657_cast_fp16 = slice_by_index(begin = var_3657_begin_0, end = var_3657_end_0, end_mask = var_3657_end_mask_0, squeeze_mask = var_3657_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3657_cast_fp16")]; + bool var_3658_transpose_x_0 = const()[name = string("op_3658_transpose_x_0"), val = bool(false)]; + bool var_3658_transpose_y_0 = const()[name = string("op_3658_transpose_y_0"), val = bool(false)]; + tensor var_3658_cast_fp16 = matmul(transpose_x = var_3658_transpose_x_0, transpose_y = var_3658_transpose_y_0, x = var_3657_cast_fp16, y = state_135_cast_fp16)[name = string("op_3658_cast_fp16")]; + tensor var_3660_begin_0 = const()[name = string("op_3660_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3660_end_0 = const()[name = string("op_3660_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_3660_end_mask_0 = const()[name = string("op_3660_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3660_squeeze_mask_0 = const()[name = string("op_3660_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3660_cast_fp16 = slice_by_index(begin = var_3660_begin_0, end = var_3660_end_0, end_mask = var_3660_end_mask_0, squeeze_mask = var_3660_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3660_cast_fp16")]; + bool var_3661_transpose_x_0 = const()[name = string("op_3661_transpose_x_0"), val = bool(false)]; + bool var_3661_transpose_y_0 = const()[name = string("op_3661_transpose_y_0"), val = bool(false)]; + tensor var_3661_cast_fp16 = matmul(transpose_x = var_3661_transpose_x_0, transpose_y = var_3661_transpose_y_0, x = var_3660_cast_fp16, y = v_new_135_cast_fp16)[name = string("op_3661_cast_fp16")]; + tensor var_3662_cast_fp16 = add(x = var_3658_cast_fp16, y = var_3661_cast_fp16)[name = string("op_3662_cast_fp16")]; + tensor var_3664_begin_0 = const()[name = string("op_3664_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3664_end_0 = const()[name = string("op_3664_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_3664_end_mask_0 = const()[name = string("op_3664_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3664_squeeze_mask_0 = const()[name = string("op_3664_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3664_cast_fp16 = slice_by_index(begin = var_3664_begin_0, end = var_3664_end_0, end_mask = var_3664_end_mask_0, squeeze_mask = var_3664_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3664_cast_fp16")]; + tensor var_3665_cast_fp16 = mul(x = state_135_cast_fp16, y = var_3664_cast_fp16)[name = string("op_3665_cast_fp16")]; + tensor var_3667_begin_0 = const()[name = string("op_3667_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_3667_end_0 = const()[name = string("op_3667_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_3667_end_mask_0 = const()[name = string("op_3667_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3667_squeeze_mask_0 = const()[name = string("op_3667_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3667_cast_fp16 = slice_by_index(begin = var_3667_begin_0, end = var_3667_end_0, end_mask = var_3667_end_mask_0, squeeze_mask = var_3667_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3667_cast_fp16")]; + bool var_3669_transpose_x_1 = const()[name = string("op_3669_transpose_x_1"), val = bool(true)]; + bool var_3669_transpose_y_1 = const()[name = string("op_3669_transpose_y_1"), val = bool(false)]; + tensor var_3669_cast_fp16 = matmul(transpose_x = var_3669_transpose_x_1, transpose_y = var_3669_transpose_y_1, x = var_3667_cast_fp16, y = v_new_135_cast_fp16)[name = string("op_3669_cast_fp16")]; + tensor state_137_cast_fp16 = add(x = var_3665_cast_fp16, y = var_3669_cast_fp16)[name = string("state_137_cast_fp16")]; + tensor var_3672_begin_0 = const()[name = string("op_3672_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3672_end_0 = const()[name = string("op_3672_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_3672_end_mask_0 = const()[name = string("op_3672_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3672_squeeze_mask_0 = const()[name = string("op_3672_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3672_cast_fp16 = slice_by_index(begin = var_3672_begin_0, end = var_3672_end_0, end_mask = var_3672_end_mask_0, squeeze_mask = var_3672_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3672_cast_fp16")]; + tensor var_3674_begin_0 = const()[name = string("op_3674_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3674_end_0 = const()[name = string("op_3674_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_3674_end_mask_0 = const()[name = string("op_3674_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3674_squeeze_mask_0 = const()[name = string("op_3674_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3674_cast_fp16 = slice_by_index(begin = var_3674_begin_0, end = var_3674_end_0, end_mask = var_3674_end_mask_0, squeeze_mask = var_3674_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3674_cast_fp16")]; + bool var_3675_transpose_x_0 = const()[name = string("op_3675_transpose_x_0"), val = bool(false)]; + bool var_3675_transpose_y_0 = const()[name = string("op_3675_transpose_y_0"), val = bool(false)]; + tensor var_3675_cast_fp16 = matmul(transpose_x = var_3675_transpose_x_0, transpose_y = var_3675_transpose_y_0, x = var_3674_cast_fp16, y = state_137_cast_fp16)[name = string("op_3675_cast_fp16")]; + tensor v_new_137_cast_fp16 = sub(x = var_3672_cast_fp16, y = var_3675_cast_fp16)[name = string("v_new_137_cast_fp16")]; + tensor var_3678_begin_0 = const()[name = string("op_3678_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3678_end_0 = const()[name = string("op_3678_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_3678_end_mask_0 = const()[name = string("op_3678_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3678_squeeze_mask_0 = const()[name = string("op_3678_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3678_cast_fp16 = slice_by_index(begin = var_3678_begin_0, end = var_3678_end_0, end_mask = var_3678_end_mask_0, squeeze_mask = var_3678_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3678_cast_fp16")]; + bool var_3679_transpose_x_0 = const()[name = string("op_3679_transpose_x_0"), val = bool(false)]; + bool var_3679_transpose_y_0 = const()[name = string("op_3679_transpose_y_0"), val = bool(false)]; + tensor var_3679_cast_fp16 = matmul(transpose_x = var_3679_transpose_x_0, transpose_y = var_3679_transpose_y_0, x = var_3678_cast_fp16, y = state_137_cast_fp16)[name = string("op_3679_cast_fp16")]; + tensor var_3681_begin_0 = const()[name = string("op_3681_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3681_end_0 = const()[name = string("op_3681_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_3681_end_mask_0 = const()[name = string("op_3681_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3681_squeeze_mask_0 = const()[name = string("op_3681_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3681_cast_fp16 = slice_by_index(begin = var_3681_begin_0, end = var_3681_end_0, end_mask = var_3681_end_mask_0, squeeze_mask = var_3681_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3681_cast_fp16")]; + bool var_3682_transpose_x_0 = const()[name = string("op_3682_transpose_x_0"), val = bool(false)]; + bool var_3682_transpose_y_0 = const()[name = string("op_3682_transpose_y_0"), val = bool(false)]; + tensor var_3682_cast_fp16 = matmul(transpose_x = var_3682_transpose_x_0, transpose_y = var_3682_transpose_y_0, x = var_3681_cast_fp16, y = v_new_137_cast_fp16)[name = string("op_3682_cast_fp16")]; + tensor var_3683_cast_fp16 = add(x = var_3679_cast_fp16, y = var_3682_cast_fp16)[name = string("op_3683_cast_fp16")]; + tensor var_3685_begin_0 = const()[name = string("op_3685_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3685_end_0 = const()[name = string("op_3685_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_3685_end_mask_0 = const()[name = string("op_3685_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3685_squeeze_mask_0 = const()[name = string("op_3685_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3685_cast_fp16 = slice_by_index(begin = var_3685_begin_0, end = var_3685_end_0, end_mask = var_3685_end_mask_0, squeeze_mask = var_3685_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3685_cast_fp16")]; + tensor var_3686_cast_fp16 = mul(x = state_137_cast_fp16, y = var_3685_cast_fp16)[name = string("op_3686_cast_fp16")]; + tensor var_3688_begin_0 = const()[name = string("op_3688_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_3688_end_0 = const()[name = string("op_3688_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_3688_end_mask_0 = const()[name = string("op_3688_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3688_squeeze_mask_0 = const()[name = string("op_3688_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3688_cast_fp16 = slice_by_index(begin = var_3688_begin_0, end = var_3688_end_0, end_mask = var_3688_end_mask_0, squeeze_mask = var_3688_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3688_cast_fp16")]; + bool var_3690_transpose_x_1 = const()[name = string("op_3690_transpose_x_1"), val = bool(true)]; + bool var_3690_transpose_y_1 = const()[name = string("op_3690_transpose_y_1"), val = bool(false)]; + tensor var_3690_cast_fp16 = matmul(transpose_x = var_3690_transpose_x_1, transpose_y = var_3690_transpose_y_1, x = var_3688_cast_fp16, y = v_new_137_cast_fp16)[name = string("op_3690_cast_fp16")]; + tensor state_139_cast_fp16 = add(x = var_3686_cast_fp16, y = var_3690_cast_fp16)[name = string("state_139_cast_fp16")]; + tensor var_3693_begin_0 = const()[name = string("op_3693_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3693_end_0 = const()[name = string("op_3693_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_3693_end_mask_0 = const()[name = string("op_3693_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3693_squeeze_mask_0 = const()[name = string("op_3693_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3693_cast_fp16 = slice_by_index(begin = var_3693_begin_0, end = var_3693_end_0, end_mask = var_3693_end_mask_0, squeeze_mask = var_3693_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3693_cast_fp16")]; + tensor var_3695_begin_0 = const()[name = string("op_3695_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3695_end_0 = const()[name = string("op_3695_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_3695_end_mask_0 = const()[name = string("op_3695_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3695_squeeze_mask_0 = const()[name = string("op_3695_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3695_cast_fp16 = slice_by_index(begin = var_3695_begin_0, end = var_3695_end_0, end_mask = var_3695_end_mask_0, squeeze_mask = var_3695_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3695_cast_fp16")]; + bool var_3696_transpose_x_0 = const()[name = string("op_3696_transpose_x_0"), val = bool(false)]; + bool var_3696_transpose_y_0 = const()[name = string("op_3696_transpose_y_0"), val = bool(false)]; + tensor var_3696_cast_fp16 = matmul(transpose_x = var_3696_transpose_x_0, transpose_y = var_3696_transpose_y_0, x = var_3695_cast_fp16, y = state_139_cast_fp16)[name = string("op_3696_cast_fp16")]; + tensor v_new_139_cast_fp16 = sub(x = var_3693_cast_fp16, y = var_3696_cast_fp16)[name = string("v_new_139_cast_fp16")]; + tensor var_3699_begin_0 = const()[name = string("op_3699_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3699_end_0 = const()[name = string("op_3699_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_3699_end_mask_0 = const()[name = string("op_3699_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3699_squeeze_mask_0 = const()[name = string("op_3699_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3699_cast_fp16 = slice_by_index(begin = var_3699_begin_0, end = var_3699_end_0, end_mask = var_3699_end_mask_0, squeeze_mask = var_3699_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3699_cast_fp16")]; + bool var_3700_transpose_x_0 = const()[name = string("op_3700_transpose_x_0"), val = bool(false)]; + bool var_3700_transpose_y_0 = const()[name = string("op_3700_transpose_y_0"), val = bool(false)]; + tensor var_3700_cast_fp16 = matmul(transpose_x = var_3700_transpose_x_0, transpose_y = var_3700_transpose_y_0, x = var_3699_cast_fp16, y = state_139_cast_fp16)[name = string("op_3700_cast_fp16")]; + tensor var_3702_begin_0 = const()[name = string("op_3702_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3702_end_0 = const()[name = string("op_3702_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_3702_end_mask_0 = const()[name = string("op_3702_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3702_squeeze_mask_0 = const()[name = string("op_3702_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3702_cast_fp16 = slice_by_index(begin = var_3702_begin_0, end = var_3702_end_0, end_mask = var_3702_end_mask_0, squeeze_mask = var_3702_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3702_cast_fp16")]; + bool var_3703_transpose_x_0 = const()[name = string("op_3703_transpose_x_0"), val = bool(false)]; + bool var_3703_transpose_y_0 = const()[name = string("op_3703_transpose_y_0"), val = bool(false)]; + tensor var_3703_cast_fp16 = matmul(transpose_x = var_3703_transpose_x_0, transpose_y = var_3703_transpose_y_0, x = var_3702_cast_fp16, y = v_new_139_cast_fp16)[name = string("op_3703_cast_fp16")]; + tensor var_3704_cast_fp16 = add(x = var_3700_cast_fp16, y = var_3703_cast_fp16)[name = string("op_3704_cast_fp16")]; + tensor var_3706_begin_0 = const()[name = string("op_3706_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3706_end_0 = const()[name = string("op_3706_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_3706_end_mask_0 = const()[name = string("op_3706_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3706_squeeze_mask_0 = const()[name = string("op_3706_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3706_cast_fp16 = slice_by_index(begin = var_3706_begin_0, end = var_3706_end_0, end_mask = var_3706_end_mask_0, squeeze_mask = var_3706_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3706_cast_fp16")]; + tensor var_3707_cast_fp16 = mul(x = state_139_cast_fp16, y = var_3706_cast_fp16)[name = string("op_3707_cast_fp16")]; + tensor var_3709_begin_0 = const()[name = string("op_3709_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_3709_end_0 = const()[name = string("op_3709_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_3709_end_mask_0 = const()[name = string("op_3709_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3709_squeeze_mask_0 = const()[name = string("op_3709_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3709_cast_fp16 = slice_by_index(begin = var_3709_begin_0, end = var_3709_end_0, end_mask = var_3709_end_mask_0, squeeze_mask = var_3709_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3709_cast_fp16")]; + bool var_3711_transpose_x_1 = const()[name = string("op_3711_transpose_x_1"), val = bool(true)]; + bool var_3711_transpose_y_1 = const()[name = string("op_3711_transpose_y_1"), val = bool(false)]; + tensor var_3711_cast_fp16 = matmul(transpose_x = var_3711_transpose_x_1, transpose_y = var_3711_transpose_y_1, x = var_3709_cast_fp16, y = v_new_139_cast_fp16)[name = string("op_3711_cast_fp16")]; + tensor state_141_cast_fp16 = add(x = var_3707_cast_fp16, y = var_3711_cast_fp16)[name = string("state_141_cast_fp16")]; + tensor var_3714_begin_0 = const()[name = string("op_3714_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3714_end_0 = const()[name = string("op_3714_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_3714_end_mask_0 = const()[name = string("op_3714_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3714_squeeze_mask_0 = const()[name = string("op_3714_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3714_cast_fp16 = slice_by_index(begin = var_3714_begin_0, end = var_3714_end_0, end_mask = var_3714_end_mask_0, squeeze_mask = var_3714_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3714_cast_fp16")]; + tensor var_3716_begin_0 = const()[name = string("op_3716_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3716_end_0 = const()[name = string("op_3716_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_3716_end_mask_0 = const()[name = string("op_3716_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3716_squeeze_mask_0 = const()[name = string("op_3716_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3716_cast_fp16 = slice_by_index(begin = var_3716_begin_0, end = var_3716_end_0, end_mask = var_3716_end_mask_0, squeeze_mask = var_3716_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3716_cast_fp16")]; + bool var_3717_transpose_x_0 = const()[name = string("op_3717_transpose_x_0"), val = bool(false)]; + bool var_3717_transpose_y_0 = const()[name = string("op_3717_transpose_y_0"), val = bool(false)]; + tensor var_3717_cast_fp16 = matmul(transpose_x = var_3717_transpose_x_0, transpose_y = var_3717_transpose_y_0, x = var_3716_cast_fp16, y = state_141_cast_fp16)[name = string("op_3717_cast_fp16")]; + tensor v_new_141_cast_fp16 = sub(x = var_3714_cast_fp16, y = var_3717_cast_fp16)[name = string("v_new_141_cast_fp16")]; + tensor var_3720_begin_0 = const()[name = string("op_3720_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3720_end_0 = const()[name = string("op_3720_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_3720_end_mask_0 = const()[name = string("op_3720_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3720_squeeze_mask_0 = const()[name = string("op_3720_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3720_cast_fp16 = slice_by_index(begin = var_3720_begin_0, end = var_3720_end_0, end_mask = var_3720_end_mask_0, squeeze_mask = var_3720_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3720_cast_fp16")]; + bool var_3721_transpose_x_0 = const()[name = string("op_3721_transpose_x_0"), val = bool(false)]; + bool var_3721_transpose_y_0 = const()[name = string("op_3721_transpose_y_0"), val = bool(false)]; + tensor var_3721_cast_fp16 = matmul(transpose_x = var_3721_transpose_x_0, transpose_y = var_3721_transpose_y_0, x = var_3720_cast_fp16, y = state_141_cast_fp16)[name = string("op_3721_cast_fp16")]; + tensor var_3723_begin_0 = const()[name = string("op_3723_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3723_end_0 = const()[name = string("op_3723_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_3723_end_mask_0 = const()[name = string("op_3723_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3723_squeeze_mask_0 = const()[name = string("op_3723_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3723_cast_fp16 = slice_by_index(begin = var_3723_begin_0, end = var_3723_end_0, end_mask = var_3723_end_mask_0, squeeze_mask = var_3723_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3723_cast_fp16")]; + bool var_3724_transpose_x_0 = const()[name = string("op_3724_transpose_x_0"), val = bool(false)]; + bool var_3724_transpose_y_0 = const()[name = string("op_3724_transpose_y_0"), val = bool(false)]; + tensor var_3724_cast_fp16 = matmul(transpose_x = var_3724_transpose_x_0, transpose_y = var_3724_transpose_y_0, x = var_3723_cast_fp16, y = v_new_141_cast_fp16)[name = string("op_3724_cast_fp16")]; + tensor var_3725_cast_fp16 = add(x = var_3721_cast_fp16, y = var_3724_cast_fp16)[name = string("op_3725_cast_fp16")]; + tensor var_3727_begin_0 = const()[name = string("op_3727_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3727_end_0 = const()[name = string("op_3727_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_3727_end_mask_0 = const()[name = string("op_3727_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3727_squeeze_mask_0 = const()[name = string("op_3727_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3727_cast_fp16 = slice_by_index(begin = var_3727_begin_0, end = var_3727_end_0, end_mask = var_3727_end_mask_0, squeeze_mask = var_3727_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3727_cast_fp16")]; + tensor var_3728_cast_fp16 = mul(x = state_141_cast_fp16, y = var_3727_cast_fp16)[name = string("op_3728_cast_fp16")]; + tensor var_3730_begin_0 = const()[name = string("op_3730_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_3730_end_0 = const()[name = string("op_3730_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_3730_end_mask_0 = const()[name = string("op_3730_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3730_squeeze_mask_0 = const()[name = string("op_3730_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3730_cast_fp16 = slice_by_index(begin = var_3730_begin_0, end = var_3730_end_0, end_mask = var_3730_end_mask_0, squeeze_mask = var_3730_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3730_cast_fp16")]; + bool var_3732_transpose_x_1 = const()[name = string("op_3732_transpose_x_1"), val = bool(true)]; + bool var_3732_transpose_y_1 = const()[name = string("op_3732_transpose_y_1"), val = bool(false)]; + tensor var_3732_cast_fp16 = matmul(transpose_x = var_3732_transpose_x_1, transpose_y = var_3732_transpose_y_1, x = var_3730_cast_fp16, y = v_new_141_cast_fp16)[name = string("op_3732_cast_fp16")]; + tensor state_143_cast_fp16 = add(x = var_3728_cast_fp16, y = var_3732_cast_fp16)[name = string("state_143_cast_fp16")]; + tensor var_3735_begin_0 = const()[name = string("op_3735_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3735_end_0 = const()[name = string("op_3735_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_3735_end_mask_0 = const()[name = string("op_3735_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3735_squeeze_mask_0 = const()[name = string("op_3735_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3735_cast_fp16 = slice_by_index(begin = var_3735_begin_0, end = var_3735_end_0, end_mask = var_3735_end_mask_0, squeeze_mask = var_3735_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3735_cast_fp16")]; + tensor var_3737_begin_0 = const()[name = string("op_3737_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3737_end_0 = const()[name = string("op_3737_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_3737_end_mask_0 = const()[name = string("op_3737_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3737_squeeze_mask_0 = const()[name = string("op_3737_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3737_cast_fp16 = slice_by_index(begin = var_3737_begin_0, end = var_3737_end_0, end_mask = var_3737_end_mask_0, squeeze_mask = var_3737_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3737_cast_fp16")]; + bool var_3738_transpose_x_0 = const()[name = string("op_3738_transpose_x_0"), val = bool(false)]; + bool var_3738_transpose_y_0 = const()[name = string("op_3738_transpose_y_0"), val = bool(false)]; + tensor var_3738_cast_fp16 = matmul(transpose_x = var_3738_transpose_x_0, transpose_y = var_3738_transpose_y_0, x = var_3737_cast_fp16, y = state_143_cast_fp16)[name = string("op_3738_cast_fp16")]; + tensor v_new_143_cast_fp16 = sub(x = var_3735_cast_fp16, y = var_3738_cast_fp16)[name = string("v_new_143_cast_fp16")]; + tensor var_3741_begin_0 = const()[name = string("op_3741_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3741_end_0 = const()[name = string("op_3741_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_3741_end_mask_0 = const()[name = string("op_3741_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3741_squeeze_mask_0 = const()[name = string("op_3741_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3741_cast_fp16 = slice_by_index(begin = var_3741_begin_0, end = var_3741_end_0, end_mask = var_3741_end_mask_0, squeeze_mask = var_3741_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3741_cast_fp16")]; + bool var_3742_transpose_x_0 = const()[name = string("op_3742_transpose_x_0"), val = bool(false)]; + bool var_3742_transpose_y_0 = const()[name = string("op_3742_transpose_y_0"), val = bool(false)]; + tensor var_3742_cast_fp16 = matmul(transpose_x = var_3742_transpose_x_0, transpose_y = var_3742_transpose_y_0, x = var_3741_cast_fp16, y = state_143_cast_fp16)[name = string("op_3742_cast_fp16")]; + tensor var_3744_begin_0 = const()[name = string("op_3744_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3744_end_0 = const()[name = string("op_3744_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_3744_end_mask_0 = const()[name = string("op_3744_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3744_squeeze_mask_0 = const()[name = string("op_3744_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3744_cast_fp16 = slice_by_index(begin = var_3744_begin_0, end = var_3744_end_0, end_mask = var_3744_end_mask_0, squeeze_mask = var_3744_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3744_cast_fp16")]; + bool var_3745_transpose_x_0 = const()[name = string("op_3745_transpose_x_0"), val = bool(false)]; + bool var_3745_transpose_y_0 = const()[name = string("op_3745_transpose_y_0"), val = bool(false)]; + tensor var_3745_cast_fp16 = matmul(transpose_x = var_3745_transpose_x_0, transpose_y = var_3745_transpose_y_0, x = var_3744_cast_fp16, y = v_new_143_cast_fp16)[name = string("op_3745_cast_fp16")]; + tensor var_3746_cast_fp16 = add(x = var_3742_cast_fp16, y = var_3745_cast_fp16)[name = string("op_3746_cast_fp16")]; + tensor var_3748_begin_0 = const()[name = string("op_3748_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3748_end_0 = const()[name = string("op_3748_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_3748_end_mask_0 = const()[name = string("op_3748_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3748_squeeze_mask_0 = const()[name = string("op_3748_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3748_cast_fp16 = slice_by_index(begin = var_3748_begin_0, end = var_3748_end_0, end_mask = var_3748_end_mask_0, squeeze_mask = var_3748_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3748_cast_fp16")]; + tensor var_3749_cast_fp16 = mul(x = state_143_cast_fp16, y = var_3748_cast_fp16)[name = string("op_3749_cast_fp16")]; + tensor var_3751_begin_0 = const()[name = string("op_3751_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_3751_end_0 = const()[name = string("op_3751_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_3751_end_mask_0 = const()[name = string("op_3751_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3751_squeeze_mask_0 = const()[name = string("op_3751_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3751_cast_fp16 = slice_by_index(begin = var_3751_begin_0, end = var_3751_end_0, end_mask = var_3751_end_mask_0, squeeze_mask = var_3751_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3751_cast_fp16")]; + bool var_3753_transpose_x_1 = const()[name = string("op_3753_transpose_x_1"), val = bool(true)]; + bool var_3753_transpose_y_1 = const()[name = string("op_3753_transpose_y_1"), val = bool(false)]; + tensor var_3753_cast_fp16 = matmul(transpose_x = var_3753_transpose_x_1, transpose_y = var_3753_transpose_y_1, x = var_3751_cast_fp16, y = v_new_143_cast_fp16)[name = string("op_3753_cast_fp16")]; + tensor state_145_cast_fp16 = add(x = var_3749_cast_fp16, y = var_3753_cast_fp16)[name = string("state_145_cast_fp16")]; + tensor var_3756_begin_0 = const()[name = string("op_3756_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3756_end_0 = const()[name = string("op_3756_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3756_end_mask_0 = const()[name = string("op_3756_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3756_squeeze_mask_0 = const()[name = string("op_3756_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3756_cast_fp16 = slice_by_index(begin = var_3756_begin_0, end = var_3756_end_0, end_mask = var_3756_end_mask_0, squeeze_mask = var_3756_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3756_cast_fp16")]; + tensor var_3758_begin_0 = const()[name = string("op_3758_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3758_end_0 = const()[name = string("op_3758_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3758_end_mask_0 = const()[name = string("op_3758_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3758_squeeze_mask_0 = const()[name = string("op_3758_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3758_cast_fp16 = slice_by_index(begin = var_3758_begin_0, end = var_3758_end_0, end_mask = var_3758_end_mask_0, squeeze_mask = var_3758_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3758_cast_fp16")]; + bool var_3759_transpose_x_0 = const()[name = string("op_3759_transpose_x_0"), val = bool(false)]; + bool var_3759_transpose_y_0 = const()[name = string("op_3759_transpose_y_0"), val = bool(false)]; + tensor var_3759_cast_fp16 = matmul(transpose_x = var_3759_transpose_x_0, transpose_y = var_3759_transpose_y_0, x = var_3758_cast_fp16, y = state_145_cast_fp16)[name = string("op_3759_cast_fp16")]; + tensor v_new_145_cast_fp16 = sub(x = var_3756_cast_fp16, y = var_3759_cast_fp16)[name = string("v_new_145_cast_fp16")]; + tensor var_3762_begin_0 = const()[name = string("op_3762_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3762_end_0 = const()[name = string("op_3762_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3762_end_mask_0 = const()[name = string("op_3762_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3762_squeeze_mask_0 = const()[name = string("op_3762_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3762_cast_fp16 = slice_by_index(begin = var_3762_begin_0, end = var_3762_end_0, end_mask = var_3762_end_mask_0, squeeze_mask = var_3762_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3762_cast_fp16")]; + bool var_3763_transpose_x_0 = const()[name = string("op_3763_transpose_x_0"), val = bool(false)]; + bool var_3763_transpose_y_0 = const()[name = string("op_3763_transpose_y_0"), val = bool(false)]; + tensor var_3763_cast_fp16 = matmul(transpose_x = var_3763_transpose_x_0, transpose_y = var_3763_transpose_y_0, x = var_3762_cast_fp16, y = state_145_cast_fp16)[name = string("op_3763_cast_fp16")]; + tensor var_3765_begin_0 = const()[name = string("op_3765_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3765_end_0 = const()[name = string("op_3765_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_3765_end_mask_0 = const()[name = string("op_3765_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3765_squeeze_mask_0 = const()[name = string("op_3765_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3765_cast_fp16 = slice_by_index(begin = var_3765_begin_0, end = var_3765_end_0, end_mask = var_3765_end_mask_0, squeeze_mask = var_3765_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3765_cast_fp16")]; + bool var_3766_transpose_x_0 = const()[name = string("op_3766_transpose_x_0"), val = bool(false)]; + bool var_3766_transpose_y_0 = const()[name = string("op_3766_transpose_y_0"), val = bool(false)]; + tensor var_3766_cast_fp16 = matmul(transpose_x = var_3766_transpose_x_0, transpose_y = var_3766_transpose_y_0, x = var_3765_cast_fp16, y = v_new_145_cast_fp16)[name = string("op_3766_cast_fp16")]; + tensor var_3767_cast_fp16 = add(x = var_3763_cast_fp16, y = var_3766_cast_fp16)[name = string("op_3767_cast_fp16")]; + tensor var_3769_begin_0 = const()[name = string("op_3769_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3769_end_0 = const()[name = string("op_3769_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_3769_end_mask_0 = const()[name = string("op_3769_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3769_squeeze_mask_0 = const()[name = string("op_3769_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3769_cast_fp16 = slice_by_index(begin = var_3769_begin_0, end = var_3769_end_0, end_mask = var_3769_end_mask_0, squeeze_mask = var_3769_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3769_cast_fp16")]; + tensor var_3770_cast_fp16 = mul(x = state_145_cast_fp16, y = var_3769_cast_fp16)[name = string("op_3770_cast_fp16")]; + tensor var_3772_begin_0 = const()[name = string("op_3772_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_3772_end_0 = const()[name = string("op_3772_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_3772_end_mask_0 = const()[name = string("op_3772_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3772_squeeze_mask_0 = const()[name = string("op_3772_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3772_cast_fp16 = slice_by_index(begin = var_3772_begin_0, end = var_3772_end_0, end_mask = var_3772_end_mask_0, squeeze_mask = var_3772_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3772_cast_fp16")]; + bool var_3774_transpose_x_1 = const()[name = string("op_3774_transpose_x_1"), val = bool(true)]; + bool var_3774_transpose_y_1 = const()[name = string("op_3774_transpose_y_1"), val = bool(false)]; + tensor var_3774_cast_fp16 = matmul(transpose_x = var_3774_transpose_x_1, transpose_y = var_3774_transpose_y_1, x = var_3772_cast_fp16, y = v_new_145_cast_fp16)[name = string("op_3774_cast_fp16")]; + tensor state_147_cast_fp16 = add(x = var_3770_cast_fp16, y = var_3774_cast_fp16)[name = string("state_147_cast_fp16")]; + tensor var_3777_begin_0 = const()[name = string("op_3777_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3777_end_0 = const()[name = string("op_3777_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3777_end_mask_0 = const()[name = string("op_3777_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3777_squeeze_mask_0 = const()[name = string("op_3777_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3777_cast_fp16 = slice_by_index(begin = var_3777_begin_0, end = var_3777_end_0, end_mask = var_3777_end_mask_0, squeeze_mask = var_3777_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3777_cast_fp16")]; + tensor var_3779_begin_0 = const()[name = string("op_3779_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3779_end_0 = const()[name = string("op_3779_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3779_end_mask_0 = const()[name = string("op_3779_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3779_squeeze_mask_0 = const()[name = string("op_3779_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3779_cast_fp16 = slice_by_index(begin = var_3779_begin_0, end = var_3779_end_0, end_mask = var_3779_end_mask_0, squeeze_mask = var_3779_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3779_cast_fp16")]; + bool var_3780_transpose_x_0 = const()[name = string("op_3780_transpose_x_0"), val = bool(false)]; + bool var_3780_transpose_y_0 = const()[name = string("op_3780_transpose_y_0"), val = bool(false)]; + tensor var_3780_cast_fp16 = matmul(transpose_x = var_3780_transpose_x_0, transpose_y = var_3780_transpose_y_0, x = var_3779_cast_fp16, y = state_147_cast_fp16)[name = string("op_3780_cast_fp16")]; + tensor v_new_147_cast_fp16 = sub(x = var_3777_cast_fp16, y = var_3780_cast_fp16)[name = string("v_new_147_cast_fp16")]; + tensor var_3783_begin_0 = const()[name = string("op_3783_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3783_end_0 = const()[name = string("op_3783_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3783_end_mask_0 = const()[name = string("op_3783_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3783_squeeze_mask_0 = const()[name = string("op_3783_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3783_cast_fp16 = slice_by_index(begin = var_3783_begin_0, end = var_3783_end_0, end_mask = var_3783_end_mask_0, squeeze_mask = var_3783_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3783_cast_fp16")]; + bool var_3784_transpose_x_0 = const()[name = string("op_3784_transpose_x_0"), val = bool(false)]; + bool var_3784_transpose_y_0 = const()[name = string("op_3784_transpose_y_0"), val = bool(false)]; + tensor var_3784_cast_fp16 = matmul(transpose_x = var_3784_transpose_x_0, transpose_y = var_3784_transpose_y_0, x = var_3783_cast_fp16, y = state_147_cast_fp16)[name = string("op_3784_cast_fp16")]; + tensor var_3786_begin_0 = const()[name = string("op_3786_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3786_end_0 = const()[name = string("op_3786_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_3786_end_mask_0 = const()[name = string("op_3786_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3786_squeeze_mask_0 = const()[name = string("op_3786_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3786_cast_fp16 = slice_by_index(begin = var_3786_begin_0, end = var_3786_end_0, end_mask = var_3786_end_mask_0, squeeze_mask = var_3786_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3786_cast_fp16")]; + bool var_3787_transpose_x_0 = const()[name = string("op_3787_transpose_x_0"), val = bool(false)]; + bool var_3787_transpose_y_0 = const()[name = string("op_3787_transpose_y_0"), val = bool(false)]; + tensor var_3787_cast_fp16 = matmul(transpose_x = var_3787_transpose_x_0, transpose_y = var_3787_transpose_y_0, x = var_3786_cast_fp16, y = v_new_147_cast_fp16)[name = string("op_3787_cast_fp16")]; + tensor var_3788_cast_fp16 = add(x = var_3784_cast_fp16, y = var_3787_cast_fp16)[name = string("op_3788_cast_fp16")]; + tensor var_3790_begin_0 = const()[name = string("op_3790_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3790_end_0 = const()[name = string("op_3790_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_3790_end_mask_0 = const()[name = string("op_3790_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3790_squeeze_mask_0 = const()[name = string("op_3790_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3790_cast_fp16 = slice_by_index(begin = var_3790_begin_0, end = var_3790_end_0, end_mask = var_3790_end_mask_0, squeeze_mask = var_3790_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3790_cast_fp16")]; + tensor var_3791_cast_fp16 = mul(x = state_147_cast_fp16, y = var_3790_cast_fp16)[name = string("op_3791_cast_fp16")]; + tensor var_3793_begin_0 = const()[name = string("op_3793_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_3793_end_0 = const()[name = string("op_3793_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_3793_end_mask_0 = const()[name = string("op_3793_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3793_squeeze_mask_0 = const()[name = string("op_3793_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3793_cast_fp16 = slice_by_index(begin = var_3793_begin_0, end = var_3793_end_0, end_mask = var_3793_end_mask_0, squeeze_mask = var_3793_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3793_cast_fp16")]; + bool var_3795_transpose_x_1 = const()[name = string("op_3795_transpose_x_1"), val = bool(true)]; + bool var_3795_transpose_y_1 = const()[name = string("op_3795_transpose_y_1"), val = bool(false)]; + tensor var_3795_cast_fp16 = matmul(transpose_x = var_3795_transpose_x_1, transpose_y = var_3795_transpose_y_1, x = var_3793_cast_fp16, y = v_new_147_cast_fp16)[name = string("op_3795_cast_fp16")]; + tensor state_149_cast_fp16 = add(x = var_3791_cast_fp16, y = var_3795_cast_fp16)[name = string("state_149_cast_fp16")]; + tensor var_3798_begin_0 = const()[name = string("op_3798_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3798_end_0 = const()[name = string("op_3798_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3798_end_mask_0 = const()[name = string("op_3798_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3798_squeeze_mask_0 = const()[name = string("op_3798_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3798_cast_fp16 = slice_by_index(begin = var_3798_begin_0, end = var_3798_end_0, end_mask = var_3798_end_mask_0, squeeze_mask = var_3798_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3798_cast_fp16")]; + tensor var_3800_begin_0 = const()[name = string("op_3800_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3800_end_0 = const()[name = string("op_3800_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3800_end_mask_0 = const()[name = string("op_3800_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3800_squeeze_mask_0 = const()[name = string("op_3800_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3800_cast_fp16 = slice_by_index(begin = var_3800_begin_0, end = var_3800_end_0, end_mask = var_3800_end_mask_0, squeeze_mask = var_3800_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3800_cast_fp16")]; + bool var_3801_transpose_x_0 = const()[name = string("op_3801_transpose_x_0"), val = bool(false)]; + bool var_3801_transpose_y_0 = const()[name = string("op_3801_transpose_y_0"), val = bool(false)]; + tensor var_3801_cast_fp16 = matmul(transpose_x = var_3801_transpose_x_0, transpose_y = var_3801_transpose_y_0, x = var_3800_cast_fp16, y = state_149_cast_fp16)[name = string("op_3801_cast_fp16")]; + tensor v_new_149_cast_fp16 = sub(x = var_3798_cast_fp16, y = var_3801_cast_fp16)[name = string("v_new_149_cast_fp16")]; + tensor var_3804_begin_0 = const()[name = string("op_3804_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3804_end_0 = const()[name = string("op_3804_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3804_end_mask_0 = const()[name = string("op_3804_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3804_squeeze_mask_0 = const()[name = string("op_3804_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3804_cast_fp16 = slice_by_index(begin = var_3804_begin_0, end = var_3804_end_0, end_mask = var_3804_end_mask_0, squeeze_mask = var_3804_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3804_cast_fp16")]; + bool var_3805_transpose_x_0 = const()[name = string("op_3805_transpose_x_0"), val = bool(false)]; + bool var_3805_transpose_y_0 = const()[name = string("op_3805_transpose_y_0"), val = bool(false)]; + tensor var_3805_cast_fp16 = matmul(transpose_x = var_3805_transpose_x_0, transpose_y = var_3805_transpose_y_0, x = var_3804_cast_fp16, y = state_149_cast_fp16)[name = string("op_3805_cast_fp16")]; + tensor var_3807_begin_0 = const()[name = string("op_3807_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3807_end_0 = const()[name = string("op_3807_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_3807_end_mask_0 = const()[name = string("op_3807_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3807_squeeze_mask_0 = const()[name = string("op_3807_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3807_cast_fp16 = slice_by_index(begin = var_3807_begin_0, end = var_3807_end_0, end_mask = var_3807_end_mask_0, squeeze_mask = var_3807_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3807_cast_fp16")]; + bool var_3808_transpose_x_0 = const()[name = string("op_3808_transpose_x_0"), val = bool(false)]; + bool var_3808_transpose_y_0 = const()[name = string("op_3808_transpose_y_0"), val = bool(false)]; + tensor var_3808_cast_fp16 = matmul(transpose_x = var_3808_transpose_x_0, transpose_y = var_3808_transpose_y_0, x = var_3807_cast_fp16, y = v_new_149_cast_fp16)[name = string("op_3808_cast_fp16")]; + tensor var_3809_cast_fp16 = add(x = var_3805_cast_fp16, y = var_3808_cast_fp16)[name = string("op_3809_cast_fp16")]; + tensor var_3811_begin_0 = const()[name = string("op_3811_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3811_end_0 = const()[name = string("op_3811_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_3811_end_mask_0 = const()[name = string("op_3811_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3811_squeeze_mask_0 = const()[name = string("op_3811_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3811_cast_fp16 = slice_by_index(begin = var_3811_begin_0, end = var_3811_end_0, end_mask = var_3811_end_mask_0, squeeze_mask = var_3811_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3811_cast_fp16")]; + tensor var_3812_cast_fp16 = mul(x = state_149_cast_fp16, y = var_3811_cast_fp16)[name = string("op_3812_cast_fp16")]; + tensor var_3814_begin_0 = const()[name = string("op_3814_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_3814_end_0 = const()[name = string("op_3814_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_3814_end_mask_0 = const()[name = string("op_3814_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3814_squeeze_mask_0 = const()[name = string("op_3814_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3814_cast_fp16 = slice_by_index(begin = var_3814_begin_0, end = var_3814_end_0, end_mask = var_3814_end_mask_0, squeeze_mask = var_3814_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3814_cast_fp16")]; + bool var_3816_transpose_x_1 = const()[name = string("op_3816_transpose_x_1"), val = bool(true)]; + bool var_3816_transpose_y_1 = const()[name = string("op_3816_transpose_y_1"), val = bool(false)]; + tensor var_3816_cast_fp16 = matmul(transpose_x = var_3816_transpose_x_1, transpose_y = var_3816_transpose_y_1, x = var_3814_cast_fp16, y = v_new_149_cast_fp16)[name = string("op_3816_cast_fp16")]; + tensor state_151_cast_fp16 = add(x = var_3812_cast_fp16, y = var_3816_cast_fp16)[name = string("state_151_cast_fp16")]; + tensor var_3819_begin_0 = const()[name = string("op_3819_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3819_end_0 = const()[name = string("op_3819_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3819_end_mask_0 = const()[name = string("op_3819_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3819_squeeze_mask_0 = const()[name = string("op_3819_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3819_cast_fp16 = slice_by_index(begin = var_3819_begin_0, end = var_3819_end_0, end_mask = var_3819_end_mask_0, squeeze_mask = var_3819_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3819_cast_fp16")]; + tensor var_3821_begin_0 = const()[name = string("op_3821_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3821_end_0 = const()[name = string("op_3821_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3821_end_mask_0 = const()[name = string("op_3821_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3821_squeeze_mask_0 = const()[name = string("op_3821_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3821_cast_fp16 = slice_by_index(begin = var_3821_begin_0, end = var_3821_end_0, end_mask = var_3821_end_mask_0, squeeze_mask = var_3821_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3821_cast_fp16")]; + bool var_3822_transpose_x_0 = const()[name = string("op_3822_transpose_x_0"), val = bool(false)]; + bool var_3822_transpose_y_0 = const()[name = string("op_3822_transpose_y_0"), val = bool(false)]; + tensor var_3822_cast_fp16 = matmul(transpose_x = var_3822_transpose_x_0, transpose_y = var_3822_transpose_y_0, x = var_3821_cast_fp16, y = state_151_cast_fp16)[name = string("op_3822_cast_fp16")]; + tensor v_new_151_cast_fp16 = sub(x = var_3819_cast_fp16, y = var_3822_cast_fp16)[name = string("v_new_151_cast_fp16")]; + tensor var_3825_begin_0 = const()[name = string("op_3825_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3825_end_0 = const()[name = string("op_3825_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3825_end_mask_0 = const()[name = string("op_3825_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3825_squeeze_mask_0 = const()[name = string("op_3825_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3825_cast_fp16 = slice_by_index(begin = var_3825_begin_0, end = var_3825_end_0, end_mask = var_3825_end_mask_0, squeeze_mask = var_3825_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3825_cast_fp16")]; + bool var_3826_transpose_x_0 = const()[name = string("op_3826_transpose_x_0"), val = bool(false)]; + bool var_3826_transpose_y_0 = const()[name = string("op_3826_transpose_y_0"), val = bool(false)]; + tensor var_3826_cast_fp16 = matmul(transpose_x = var_3826_transpose_x_0, transpose_y = var_3826_transpose_y_0, x = var_3825_cast_fp16, y = state_151_cast_fp16)[name = string("op_3826_cast_fp16")]; + tensor var_3828_begin_0 = const()[name = string("op_3828_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3828_end_0 = const()[name = string("op_3828_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_3828_end_mask_0 = const()[name = string("op_3828_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3828_squeeze_mask_0 = const()[name = string("op_3828_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3828_cast_fp16 = slice_by_index(begin = var_3828_begin_0, end = var_3828_end_0, end_mask = var_3828_end_mask_0, squeeze_mask = var_3828_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3828_cast_fp16")]; + bool var_3829_transpose_x_0 = const()[name = string("op_3829_transpose_x_0"), val = bool(false)]; + bool var_3829_transpose_y_0 = const()[name = string("op_3829_transpose_y_0"), val = bool(false)]; + tensor var_3829_cast_fp16 = matmul(transpose_x = var_3829_transpose_x_0, transpose_y = var_3829_transpose_y_0, x = var_3828_cast_fp16, y = v_new_151_cast_fp16)[name = string("op_3829_cast_fp16")]; + tensor var_3830_cast_fp16 = add(x = var_3826_cast_fp16, y = var_3829_cast_fp16)[name = string("op_3830_cast_fp16")]; + tensor var_3832_begin_0 = const()[name = string("op_3832_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3832_end_0 = const()[name = string("op_3832_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_3832_end_mask_0 = const()[name = string("op_3832_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3832_squeeze_mask_0 = const()[name = string("op_3832_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3832_cast_fp16 = slice_by_index(begin = var_3832_begin_0, end = var_3832_end_0, end_mask = var_3832_end_mask_0, squeeze_mask = var_3832_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3832_cast_fp16")]; + tensor var_3833_cast_fp16 = mul(x = state_151_cast_fp16, y = var_3832_cast_fp16)[name = string("op_3833_cast_fp16")]; + tensor var_3835_begin_0 = const()[name = string("op_3835_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_3835_end_0 = const()[name = string("op_3835_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_3835_end_mask_0 = const()[name = string("op_3835_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3835_squeeze_mask_0 = const()[name = string("op_3835_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3835_cast_fp16 = slice_by_index(begin = var_3835_begin_0, end = var_3835_end_0, end_mask = var_3835_end_mask_0, squeeze_mask = var_3835_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3835_cast_fp16")]; + bool var_3837_transpose_x_1 = const()[name = string("op_3837_transpose_x_1"), val = bool(true)]; + bool var_3837_transpose_y_1 = const()[name = string("op_3837_transpose_y_1"), val = bool(false)]; + tensor var_3837_cast_fp16 = matmul(transpose_x = var_3837_transpose_x_1, transpose_y = var_3837_transpose_y_1, x = var_3835_cast_fp16, y = v_new_151_cast_fp16)[name = string("op_3837_cast_fp16")]; + tensor state_153_cast_fp16 = add(x = var_3833_cast_fp16, y = var_3837_cast_fp16)[name = string("state_153_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = string("op_3840_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3840_end_0 = const()[name = string("op_3840_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3840_end_mask_0 = const()[name = string("op_3840_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3840_squeeze_mask_0 = const()[name = string("op_3840_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, squeeze_mask = var_3840_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3840_cast_fp16")]; + tensor var_3842_begin_0 = const()[name = string("op_3842_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3842_end_0 = const()[name = string("op_3842_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3842_end_mask_0 = const()[name = string("op_3842_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3842_squeeze_mask_0 = const()[name = string("op_3842_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3842_cast_fp16 = slice_by_index(begin = var_3842_begin_0, end = var_3842_end_0, end_mask = var_3842_end_mask_0, squeeze_mask = var_3842_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3842_cast_fp16")]; + bool var_3843_transpose_x_0 = const()[name = string("op_3843_transpose_x_0"), val = bool(false)]; + bool var_3843_transpose_y_0 = const()[name = string("op_3843_transpose_y_0"), val = bool(false)]; + tensor var_3843_cast_fp16 = matmul(transpose_x = var_3843_transpose_x_0, transpose_y = var_3843_transpose_y_0, x = var_3842_cast_fp16, y = state_153_cast_fp16)[name = string("op_3843_cast_fp16")]; + tensor v_new_153_cast_fp16 = sub(x = var_3840_cast_fp16, y = var_3843_cast_fp16)[name = string("v_new_153_cast_fp16")]; + tensor var_3846_begin_0 = const()[name = string("op_3846_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3846_end_0 = const()[name = string("op_3846_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3846_end_mask_0 = const()[name = string("op_3846_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3846_squeeze_mask_0 = const()[name = string("op_3846_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3846_cast_fp16 = slice_by_index(begin = var_3846_begin_0, end = var_3846_end_0, end_mask = var_3846_end_mask_0, squeeze_mask = var_3846_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3846_cast_fp16")]; + bool var_3847_transpose_x_0 = const()[name = string("op_3847_transpose_x_0"), val = bool(false)]; + bool var_3847_transpose_y_0 = const()[name = string("op_3847_transpose_y_0"), val = bool(false)]; + tensor var_3847_cast_fp16 = matmul(transpose_x = var_3847_transpose_x_0, transpose_y = var_3847_transpose_y_0, x = var_3846_cast_fp16, y = state_153_cast_fp16)[name = string("op_3847_cast_fp16")]; + tensor var_3849_begin_0 = const()[name = string("op_3849_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3849_end_0 = const()[name = string("op_3849_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_3849_end_mask_0 = const()[name = string("op_3849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3849_squeeze_mask_0 = const()[name = string("op_3849_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3849_cast_fp16 = slice_by_index(begin = var_3849_begin_0, end = var_3849_end_0, end_mask = var_3849_end_mask_0, squeeze_mask = var_3849_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3849_cast_fp16")]; + bool var_3850_transpose_x_0 = const()[name = string("op_3850_transpose_x_0"), val = bool(false)]; + bool var_3850_transpose_y_0 = const()[name = string("op_3850_transpose_y_0"), val = bool(false)]; + tensor var_3850_cast_fp16 = matmul(transpose_x = var_3850_transpose_x_0, transpose_y = var_3850_transpose_y_0, x = var_3849_cast_fp16, y = v_new_153_cast_fp16)[name = string("op_3850_cast_fp16")]; + tensor var_3851_cast_fp16 = add(x = var_3847_cast_fp16, y = var_3850_cast_fp16)[name = string("op_3851_cast_fp16")]; + tensor var_3853_begin_0 = const()[name = string("op_3853_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3853_end_0 = const()[name = string("op_3853_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_3853_end_mask_0 = const()[name = string("op_3853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3853_squeeze_mask_0 = const()[name = string("op_3853_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3853_cast_fp16 = slice_by_index(begin = var_3853_begin_0, end = var_3853_end_0, end_mask = var_3853_end_mask_0, squeeze_mask = var_3853_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3853_cast_fp16")]; + tensor var_3854_cast_fp16 = mul(x = state_153_cast_fp16, y = var_3853_cast_fp16)[name = string("op_3854_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = string("op_3856_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_3856_end_0 = const()[name = string("op_3856_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_3856_end_mask_0 = const()[name = string("op_3856_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3856_squeeze_mask_0 = const()[name = string("op_3856_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, squeeze_mask = var_3856_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3856_cast_fp16")]; + bool var_3858_transpose_x_1 = const()[name = string("op_3858_transpose_x_1"), val = bool(true)]; + bool var_3858_transpose_y_1 = const()[name = string("op_3858_transpose_y_1"), val = bool(false)]; + tensor var_3858_cast_fp16 = matmul(transpose_x = var_3858_transpose_x_1, transpose_y = var_3858_transpose_y_1, x = var_3856_cast_fp16, y = v_new_153_cast_fp16)[name = string("op_3858_cast_fp16")]; + tensor state_155_cast_fp16 = add(x = var_3854_cast_fp16, y = var_3858_cast_fp16)[name = string("state_155_cast_fp16")]; + tensor var_3861_begin_0 = const()[name = string("op_3861_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3861_end_0 = const()[name = string("op_3861_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3861_end_mask_0 = const()[name = string("op_3861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3861_squeeze_mask_0 = const()[name = string("op_3861_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3861_cast_fp16 = slice_by_index(begin = var_3861_begin_0, end = var_3861_end_0, end_mask = var_3861_end_mask_0, squeeze_mask = var_3861_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3861_cast_fp16")]; + tensor var_3863_begin_0 = const()[name = string("op_3863_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3863_end_0 = const()[name = string("op_3863_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3863_end_mask_0 = const()[name = string("op_3863_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3863_squeeze_mask_0 = const()[name = string("op_3863_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3863_cast_fp16 = slice_by_index(begin = var_3863_begin_0, end = var_3863_end_0, end_mask = var_3863_end_mask_0, squeeze_mask = var_3863_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3863_cast_fp16")]; + bool var_3864_transpose_x_0 = const()[name = string("op_3864_transpose_x_0"), val = bool(false)]; + bool var_3864_transpose_y_0 = const()[name = string("op_3864_transpose_y_0"), val = bool(false)]; + tensor var_3864_cast_fp16 = matmul(transpose_x = var_3864_transpose_x_0, transpose_y = var_3864_transpose_y_0, x = var_3863_cast_fp16, y = state_155_cast_fp16)[name = string("op_3864_cast_fp16")]; + tensor v_new_155_cast_fp16 = sub(x = var_3861_cast_fp16, y = var_3864_cast_fp16)[name = string("v_new_155_cast_fp16")]; + tensor var_3867_begin_0 = const()[name = string("op_3867_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3867_end_0 = const()[name = string("op_3867_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3867_end_mask_0 = const()[name = string("op_3867_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3867_squeeze_mask_0 = const()[name = string("op_3867_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3867_cast_fp16 = slice_by_index(begin = var_3867_begin_0, end = var_3867_end_0, end_mask = var_3867_end_mask_0, squeeze_mask = var_3867_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3867_cast_fp16")]; + bool var_3868_transpose_x_0 = const()[name = string("op_3868_transpose_x_0"), val = bool(false)]; + bool var_3868_transpose_y_0 = const()[name = string("op_3868_transpose_y_0"), val = bool(false)]; + tensor var_3868_cast_fp16 = matmul(transpose_x = var_3868_transpose_x_0, transpose_y = var_3868_transpose_y_0, x = var_3867_cast_fp16, y = state_155_cast_fp16)[name = string("op_3868_cast_fp16")]; + tensor var_3870_begin_0 = const()[name = string("op_3870_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3870_end_0 = const()[name = string("op_3870_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_3870_end_mask_0 = const()[name = string("op_3870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3870_squeeze_mask_0 = const()[name = string("op_3870_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3870_cast_fp16 = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, squeeze_mask = var_3870_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3870_cast_fp16")]; + bool var_3871_transpose_x_0 = const()[name = string("op_3871_transpose_x_0"), val = bool(false)]; + bool var_3871_transpose_y_0 = const()[name = string("op_3871_transpose_y_0"), val = bool(false)]; + tensor var_3871_cast_fp16 = matmul(transpose_x = var_3871_transpose_x_0, transpose_y = var_3871_transpose_y_0, x = var_3870_cast_fp16, y = v_new_155_cast_fp16)[name = string("op_3871_cast_fp16")]; + tensor var_3872_cast_fp16 = add(x = var_3868_cast_fp16, y = var_3871_cast_fp16)[name = string("op_3872_cast_fp16")]; + tensor var_3874_begin_0 = const()[name = string("op_3874_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3874_end_0 = const()[name = string("op_3874_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_3874_end_mask_0 = const()[name = string("op_3874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3874_squeeze_mask_0 = const()[name = string("op_3874_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3874_cast_fp16 = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, squeeze_mask = var_3874_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3874_cast_fp16")]; + tensor var_3875_cast_fp16 = mul(x = state_155_cast_fp16, y = var_3874_cast_fp16)[name = string("op_3875_cast_fp16")]; + tensor var_3877_begin_0 = const()[name = string("op_3877_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_3877_end_0 = const()[name = string("op_3877_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_3877_end_mask_0 = const()[name = string("op_3877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3877_squeeze_mask_0 = const()[name = string("op_3877_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3877_cast_fp16 = slice_by_index(begin = var_3877_begin_0, end = var_3877_end_0, end_mask = var_3877_end_mask_0, squeeze_mask = var_3877_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3877_cast_fp16")]; + bool var_3879_transpose_x_1 = const()[name = string("op_3879_transpose_x_1"), val = bool(true)]; + bool var_3879_transpose_y_1 = const()[name = string("op_3879_transpose_y_1"), val = bool(false)]; + tensor var_3879_cast_fp16 = matmul(transpose_x = var_3879_transpose_x_1, transpose_y = var_3879_transpose_y_1, x = var_3877_cast_fp16, y = v_new_155_cast_fp16)[name = string("op_3879_cast_fp16")]; + tensor state_157_cast_fp16 = add(x = var_3875_cast_fp16, y = var_3879_cast_fp16)[name = string("state_157_cast_fp16")]; + tensor var_3882_begin_0 = const()[name = string("op_3882_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3882_end_0 = const()[name = string("op_3882_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3882_end_mask_0 = const()[name = string("op_3882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3882_squeeze_mask_0 = const()[name = string("op_3882_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3882_cast_fp16 = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, squeeze_mask = var_3882_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3882_cast_fp16")]; + tensor var_3884_begin_0 = const()[name = string("op_3884_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3884_end_0 = const()[name = string("op_3884_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3884_end_mask_0 = const()[name = string("op_3884_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3884_squeeze_mask_0 = const()[name = string("op_3884_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3884_cast_fp16 = slice_by_index(begin = var_3884_begin_0, end = var_3884_end_0, end_mask = var_3884_end_mask_0, squeeze_mask = var_3884_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3884_cast_fp16")]; + bool var_3885_transpose_x_0 = const()[name = string("op_3885_transpose_x_0"), val = bool(false)]; + bool var_3885_transpose_y_0 = const()[name = string("op_3885_transpose_y_0"), val = bool(false)]; + tensor var_3885_cast_fp16 = matmul(transpose_x = var_3885_transpose_x_0, transpose_y = var_3885_transpose_y_0, x = var_3884_cast_fp16, y = state_157_cast_fp16)[name = string("op_3885_cast_fp16")]; + tensor v_new_157_cast_fp16 = sub(x = var_3882_cast_fp16, y = var_3885_cast_fp16)[name = string("v_new_157_cast_fp16")]; + tensor var_3888_begin_0 = const()[name = string("op_3888_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3888_end_0 = const()[name = string("op_3888_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3888_end_mask_0 = const()[name = string("op_3888_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3888_squeeze_mask_0 = const()[name = string("op_3888_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3888_cast_fp16 = slice_by_index(begin = var_3888_begin_0, end = var_3888_end_0, end_mask = var_3888_end_mask_0, squeeze_mask = var_3888_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3888_cast_fp16")]; + bool var_3889_transpose_x_0 = const()[name = string("op_3889_transpose_x_0"), val = bool(false)]; + bool var_3889_transpose_y_0 = const()[name = string("op_3889_transpose_y_0"), val = bool(false)]; + tensor var_3889_cast_fp16 = matmul(transpose_x = var_3889_transpose_x_0, transpose_y = var_3889_transpose_y_0, x = var_3888_cast_fp16, y = state_157_cast_fp16)[name = string("op_3889_cast_fp16")]; + tensor var_3891_begin_0 = const()[name = string("op_3891_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3891_end_0 = const()[name = string("op_3891_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_3891_end_mask_0 = const()[name = string("op_3891_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3891_squeeze_mask_0 = const()[name = string("op_3891_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3891_cast_fp16 = slice_by_index(begin = var_3891_begin_0, end = var_3891_end_0, end_mask = var_3891_end_mask_0, squeeze_mask = var_3891_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3891_cast_fp16")]; + bool var_3892_transpose_x_0 = const()[name = string("op_3892_transpose_x_0"), val = bool(false)]; + bool var_3892_transpose_y_0 = const()[name = string("op_3892_transpose_y_0"), val = bool(false)]; + tensor var_3892_cast_fp16 = matmul(transpose_x = var_3892_transpose_x_0, transpose_y = var_3892_transpose_y_0, x = var_3891_cast_fp16, y = v_new_157_cast_fp16)[name = string("op_3892_cast_fp16")]; + tensor var_3893_cast_fp16 = add(x = var_3889_cast_fp16, y = var_3892_cast_fp16)[name = string("op_3893_cast_fp16")]; + tensor var_3895_begin_0 = const()[name = string("op_3895_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3895_end_0 = const()[name = string("op_3895_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_3895_end_mask_0 = const()[name = string("op_3895_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3895_squeeze_mask_0 = const()[name = string("op_3895_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3895_cast_fp16 = slice_by_index(begin = var_3895_begin_0, end = var_3895_end_0, end_mask = var_3895_end_mask_0, squeeze_mask = var_3895_squeeze_mask_0, x = chunk_decay_9_cast_fp16)[name = string("op_3895_cast_fp16")]; + tensor var_3896_cast_fp16 = mul(x = state_157_cast_fp16, y = var_3895_cast_fp16)[name = string("op_3896_cast_fp16")]; + tensor var_3898_begin_0 = const()[name = string("op_3898_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_3898_end_0 = const()[name = string("op_3898_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_3898_end_mask_0 = const()[name = string("op_3898_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3898_squeeze_mask_0 = const()[name = string("op_3898_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3898_cast_fp16 = slice_by_index(begin = var_3898_begin_0, end = var_3898_end_0, end_mask = var_3898_end_mask_0, squeeze_mask = var_3898_squeeze_mask_0, x = k_dec_9_cast_fp16)[name = string("op_3898_cast_fp16")]; + bool var_3900_transpose_x_1 = const()[name = string("op_3900_transpose_x_1"), val = bool(true)]; + bool var_3900_transpose_y_1 = const()[name = string("op_3900_transpose_y_1"), val = bool(false)]; + tensor var_3900_cast_fp16 = matmul(transpose_x = var_3900_transpose_x_1, transpose_y = var_3900_transpose_y_1, x = var_3898_cast_fp16, y = v_new_157_cast_fp16)[name = string("op_3900_cast_fp16")]; + tensor state_159_cast_fp16 = add(x = var_3896_cast_fp16, y = var_3900_cast_fp16)[name = string("state_159_cast_fp16")]; + tensor var_3903_begin_0 = const()[name = string("op_3903_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3903_end_0 = const()[name = string("op_3903_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3903_end_mask_0 = const()[name = string("op_3903_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3903_squeeze_mask_0 = const()[name = string("op_3903_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3903_cast_fp16 = slice_by_index(begin = var_3903_begin_0, end = var_3903_end_0, end_mask = var_3903_end_mask_0, squeeze_mask = var_3903_squeeze_mask_0, x = new_v_9_cast_fp16)[name = string("op_3903_cast_fp16")]; + tensor var_3905_begin_0 = const()[name = string("op_3905_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3905_end_0 = const()[name = string("op_3905_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3905_end_mask_0 = const()[name = string("op_3905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3905_squeeze_mask_0 = const()[name = string("op_3905_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3905_cast_fp16 = slice_by_index(begin = var_3905_begin_0, end = var_3905_end_0, end_mask = var_3905_end_mask_0, squeeze_mask = var_3905_squeeze_mask_0, x = k_cumdecay_9_cast_fp16)[name = string("op_3905_cast_fp16")]; + bool var_3906_transpose_x_0 = const()[name = string("op_3906_transpose_x_0"), val = bool(false)]; + bool var_3906_transpose_y_0 = const()[name = string("op_3906_transpose_y_0"), val = bool(false)]; + tensor var_3906_cast_fp16 = matmul(transpose_x = var_3906_transpose_x_0, transpose_y = var_3906_transpose_y_0, x = var_3905_cast_fp16, y = state_159_cast_fp16)[name = string("op_3906_cast_fp16")]; + tensor v_new_159_cast_fp16 = sub(x = var_3903_cast_fp16, y = var_3906_cast_fp16)[name = string("v_new_159_cast_fp16")]; + tensor var_3909_begin_0 = const()[name = string("op_3909_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3909_end_0 = const()[name = string("op_3909_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_3909_end_mask_0 = const()[name = string("op_3909_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3909_squeeze_mask_0 = const()[name = string("op_3909_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3909_cast_fp16 = slice_by_index(begin = var_3909_begin_0, end = var_3909_end_0, end_mask = var_3909_end_mask_0, squeeze_mask = var_3909_squeeze_mask_0, x = q_dec_9_cast_fp16)[name = string("op_3909_cast_fp16")]; + bool var_3910_transpose_x_0 = const()[name = string("op_3910_transpose_x_0"), val = bool(false)]; + bool var_3910_transpose_y_0 = const()[name = string("op_3910_transpose_y_0"), val = bool(false)]; + tensor var_3910_cast_fp16 = matmul(transpose_x = var_3910_transpose_x_0, transpose_y = var_3910_transpose_y_0, x = var_3909_cast_fp16, y = state_159_cast_fp16)[name = string("op_3910_cast_fp16")]; + tensor var_3912_begin_0 = const()[name = string("op_3912_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_3912_end_0 = const()[name = string("op_3912_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_3912_end_mask_0 = const()[name = string("op_3912_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3912_squeeze_mask_0 = const()[name = string("op_3912_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_3912_cast_fp16 = slice_by_index(begin = var_3912_begin_0, end = var_3912_end_0, end_mask = var_3912_end_mask_0, squeeze_mask = var_3912_squeeze_mask_0, x = attn_intra_9_cast_fp16)[name = string("op_3912_cast_fp16")]; + bool var_3913_transpose_x_0 = const()[name = string("op_3913_transpose_x_0"), val = bool(false)]; + bool var_3913_transpose_y_0 = const()[name = string("op_3913_transpose_y_0"), val = bool(false)]; + tensor var_3913_cast_fp16 = matmul(transpose_x = var_3913_transpose_x_0, transpose_y = var_3913_transpose_y_0, x = var_3912_cast_fp16, y = v_new_159_cast_fp16)[name = string("op_3913_cast_fp16")]; + tensor var_3914_cast_fp16 = add(x = var_3910_cast_fp16, y = var_3913_cast_fp16)[name = string("op_3914_cast_fp16")]; + int32 var_3916_axis_0 = const()[name = string("op_3916_axis_0"), val = int32(1)]; + tensor var_3916_cast_fp16 = stack(axis = var_3916_axis_0, values = (var_3599_cast_fp16, var_3620_cast_fp16, var_3641_cast_fp16, var_3662_cast_fp16, var_3683_cast_fp16, var_3704_cast_fp16, var_3725_cast_fp16, var_3746_cast_fp16, var_3767_cast_fp16, var_3788_cast_fp16, var_3809_cast_fp16, var_3830_cast_fp16, var_3851_cast_fp16, var_3872_cast_fp16, var_3893_cast_fp16, var_3914_cast_fp16))[name = string("op_3916_cast_fp16")]; + tensor var_3917 = const()[name = string("op_3917"), val = tensor([32, 1024, 128])]; + tensor var_3918_cast_fp16 = reshape(shape = var_3917, x = var_3916_cast_fp16)[name = string("op_3918_cast_fp16")]; + tensor var_3919 = const()[name = string("op_3919"), val = tensor([1, 0, 2])]; + tensor input_79_begin_0 = const()[name = string("input_79_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_79_end_0 = const()[name = string("input_79_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_79_end_mask_0 = const()[name = string("input_79_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_79_squeeze_mask_0 = const()[name = string("input_79_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_79_cast_fp16 = slice_by_index(begin = input_79_begin_0, end = input_79_end_0, end_mask = input_79_end_mask_0, squeeze_mask = input_79_squeeze_mask_0, x = z_9_cast_fp16)[name = string("input_79_cast_fp16")]; + fp16 var_3241_promoted_1_to_fp16 = const()[name = string("op_3241_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_143_cast_fp16 = transpose(perm = var_3919, x = var_3918_cast_fp16)[name = string("transpose_12")]; + tensor var_3923_cast_fp16 = pow(x = x_143_cast_fp16, y = var_3241_promoted_1_to_fp16)[name = string("op_3923_cast_fp16")]; + tensor var_3925_axes_0 = const()[name = string("op_3925_axes_0"), val = tensor([-1])]; + bool var_3925_keep_dims_0 = const()[name = string("op_3925_keep_dims_0"), val = bool(true)]; + tensor var_3925_cast_fp16 = reduce_mean(axes = var_3925_axes_0, keep_dims = var_3925_keep_dims_0, x = var_3923_cast_fp16)[name = string("op_3925_cast_fp16")]; + fp16 var_3926_to_fp16 = const()[name = string("op_3926_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3927_cast_fp16 = add(x = var_3925_cast_fp16, y = var_3926_to_fp16)[name = string("op_3927_cast_fp16")]; + fp32 var_3928_epsilon_0 = const()[name = string("op_3928_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3928_cast_fp16 = rsqrt(epsilon = var_3928_epsilon_0, x = var_3927_cast_fp16)[name = string("op_3928_cast_fp16")]; + tensor x_145_cast_fp16 = mul(x = x_143_cast_fp16, y = var_3928_cast_fp16)[name = string("x_145_cast_fp16")]; + tensor layers_5_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_5_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187566976)))]; + tensor var_3930_cast_fp16 = mul(x = layers_5_linear_attn_norm_weight_to_fp16, y = x_145_cast_fp16)[name = string("op_3930_cast_fp16")]; + tensor var_3931_cast_fp16 = silu(x = input_79_cast_fp16)[name = string("op_3931_cast_fp16")]; + tensor core_9_cast_fp16 = mul(x = var_3930_cast_fp16, y = var_3931_cast_fp16)[name = string("core_9_cast_fp16")]; + tensor var_3933 = const()[name = string("op_3933"), val = tensor([1, 1024, 4096])]; + tensor input_81_cast_fp16 = reshape(shape = var_3933, x = core_9_cast_fp16)[name = string("input_81_cast_fp16")]; + tensor layers_5_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_5_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1187567296)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_5_linear_attn_out_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_43_cast_fp16")]; + tensor x_147_cast_fp16 = add(x = x_125_cast_fp16, y = linear_43_cast_fp16)[name = string("x_147_cast_fp16")]; + fp16 var_3241_promoted_2_to_fp16 = const()[name = string("op_3241_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_3939_cast_fp16 = pow(x = x_147_cast_fp16, y = var_3241_promoted_2_to_fp16)[name = string("op_3939_cast_fp16")]; + tensor var_3941_axes_0 = const()[name = string("op_3941_axes_0"), val = tensor([-1])]; + bool var_3941_keep_dims_0 = const()[name = string("op_3941_keep_dims_0"), val = bool(true)]; + tensor var_3941_cast_fp16 = reduce_mean(axes = var_3941_axes_0, keep_dims = var_3941_keep_dims_0, x = var_3939_cast_fp16)[name = string("op_3941_cast_fp16")]; + fp16 var_3942_to_fp16 = const()[name = string("op_3942_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_3943_cast_fp16 = add(x = var_3941_cast_fp16, y = var_3942_to_fp16)[name = string("op_3943_cast_fp16")]; + fp32 var_3944_epsilon_0 = const()[name = string("op_3944_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_3944_cast_fp16 = rsqrt(epsilon = var_3944_epsilon_0, x = var_3943_cast_fp16)[name = string("op_3944_cast_fp16")]; + tensor x_149_cast_fp16 = mul(x = x_147_cast_fp16, y = var_3944_cast_fp16)[name = string("x_149_cast_fp16")]; + tensor var_3947_to_fp16 = const()[name = string("op_3947_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208538880)))]; + tensor input_83_cast_fp16 = mul(x = x_149_cast_fp16, y = var_3947_to_fp16)[name = string("input_83_cast_fp16")]; + tensor layers_5_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1208544064)))]; + tensor linear_44_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_5_mlp_gate_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_44_cast_fp16")]; + tensor var_3954_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("op_3954_cast_fp16")]; + tensor layers_5_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1255730048)))]; + tensor linear_45_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_5_mlp_up_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_45_cast_fp16")]; + tensor input_87_cast_fp16 = mul(x = var_3954_cast_fp16, y = linear_45_cast_fp16)[name = string("input_87_cast_fp16")]; + tensor layers_5_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1302916032)))]; + tensor linear_46_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_5_mlp_down_proj_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_46_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_147_cast_fp16, y = linear_46_cast_fp16)[name = string("x_151_cast_fp16")]; + int32 var_3974 = const()[name = string("op_3974"), val = int32(-2)]; + int32 var_3995 = const()[name = string("op_3995"), val = int32(-1)]; + fp16 var_3994_promoted_to_fp16 = const()[name = string("op_3994_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4004_cast_fp16 = pow(x = x_151_cast_fp16, y = var_3994_promoted_to_fp16)[name = string("op_4004_cast_fp16")]; + tensor var_4006_axes_0 = const()[name = string("op_4006_axes_0"), val = tensor([-1])]; + bool var_4006_keep_dims_0 = const()[name = string("op_4006_keep_dims_0"), val = bool(true)]; + tensor var_4006_cast_fp16 = reduce_mean(axes = var_4006_axes_0, keep_dims = var_4006_keep_dims_0, x = var_4004_cast_fp16)[name = string("op_4006_cast_fp16")]; + fp16 var_4007_to_fp16 = const()[name = string("op_4007_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4008_cast_fp16 = add(x = var_4006_cast_fp16, y = var_4007_to_fp16)[name = string("op_4008_cast_fp16")]; + fp32 var_4009_epsilon_0 = const()[name = string("op_4009_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4009_cast_fp16 = rsqrt(epsilon = var_4009_epsilon_0, x = var_4008_cast_fp16)[name = string("op_4009_cast_fp16")]; + tensor x_153_cast_fp16 = mul(x = x_151_cast_fp16, y = var_4009_cast_fp16)[name = string("x_153_cast_fp16")]; + tensor var_4012_to_fp16 = const()[name = string("op_4012_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1350102016)))]; + tensor x_155_cast_fp16 = mul(x = x_153_cast_fp16, y = var_4012_to_fp16)[name = string("x_155_cast_fp16")]; + tensor layers_6_linear_attn_in_proj_qkv_weight_to_fp16 = const()[name = string("layers_6_linear_attn_in_proj_qkv_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1350107200)))]; + tensor linear_47_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_linear_attn_in_proj_qkv_weight_to_fp16, x = x_155_cast_fp16)[name = string("linear_47_cast_fp16")]; + tensor input_89_perm_0 = const()[name = string("input_89_perm_0"), val = tensor([0, 2, 1])]; + tensor input_91_pad_0 = const()[name = string("input_91_pad_0"), val = tensor([0, 0, 0, 0, 3, 0])]; + string input_91_mode_0 = const()[name = string("input_91_mode_0"), val = string("constant")]; + fp16 const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = fp16(0x0p+0)]; + tensor input_89_cast_fp16 = transpose(perm = input_89_perm_0, x = linear_47_cast_fp16)[name = string("transpose_11")]; + tensor input_91_cast_fp16 = pad(constant_val = const_66_to_fp16, mode = input_91_mode_0, pad = input_91_pad_0, x = input_89_cast_fp16)[name = string("input_91_cast_fp16")]; + string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")]; + int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(8192)]; + tensor input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor([1])]; + tensor input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor([0, 0])]; + tensor input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor([1])]; + tensor layers_6_linear_attn_conv1d_weight_to_fp16 = const()[name = string("layers_6_linear_attn_conv1d_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1392050304)))]; + tensor input_93_cast_fp16 = conv(dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_6_linear_attn_conv1d_weight_to_fp16, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")]; + tensor var_4041_cast_fp16 = silu(x = input_93_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor qkv_perm_0 = const()[name = string("qkv_perm_0"), val = tensor([0, 2, 1])]; + tensor var_4043_begin_0 = const()[name = string("op_4043_begin_0"), val = tensor([0, 0, 0])]; + tensor var_4043_end_0 = const()[name = string("op_4043_end_0"), val = tensor([1, 1024, 2048])]; + tensor var_4043_end_mask_0 = const()[name = string("op_4043_end_mask_0"), val = tensor([true, true, false])]; + tensor qkv_cast_fp16 = transpose(perm = qkv_perm_0, x = var_4041_cast_fp16)[name = string("transpose_10")]; + tensor var_4043_cast_fp16 = slice_by_index(begin = var_4043_begin_0, end = var_4043_end_0, end_mask = var_4043_end_mask_0, x = qkv_cast_fp16)[name = string("op_4043_cast_fp16")]; + tensor var_4044 = const()[name = string("op_4044"), val = tensor([1, 1024, 16, 128])]; + tensor q_55_cast_fp16 = reshape(shape = var_4044, x = var_4043_cast_fp16)[name = string("q_55_cast_fp16")]; + tensor var_4046_begin_0 = const()[name = string("op_4046_begin_0"), val = tensor([0, 0, 2048])]; + tensor var_4046_end_0 = const()[name = string("op_4046_end_0"), val = tensor([1, 1024, 4096])]; + tensor var_4046_end_mask_0 = const()[name = string("op_4046_end_mask_0"), val = tensor([true, true, false])]; + tensor var_4046_cast_fp16 = slice_by_index(begin = var_4046_begin_0, end = var_4046_end_0, end_mask = var_4046_end_mask_0, x = qkv_cast_fp16)[name = string("op_4046_cast_fp16")]; + tensor var_4047 = const()[name = string("op_4047"), val = tensor([1, 1024, 16, 128])]; + tensor k_47_cast_fp16 = reshape(shape = var_4047, x = var_4046_cast_fp16)[name = string("k_47_cast_fp16")]; + tensor var_4049_begin_0 = const()[name = string("op_4049_begin_0"), val = tensor([0, 0, 4096])]; + tensor var_4049_end_0 = const()[name = string("op_4049_end_0"), val = tensor([1, 1024, 8192])]; + tensor var_4049_end_mask_0 = const()[name = string("op_4049_end_mask_0"), val = tensor([true, true, true])]; + tensor var_4049_cast_fp16 = slice_by_index(begin = var_4049_begin_0, end = var_4049_end_0, end_mask = var_4049_end_mask_0, x = qkv_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4050 = const()[name = string("op_4050"), val = tensor([1, 1024, 32, 128])]; + tensor v_25_cast_fp16 = reshape(shape = var_4050, x = var_4049_cast_fp16)[name = string("v_25_cast_fp16")]; + tensor layers_6_linear_attn_in_proj_z_weight_to_fp16 = const()[name = string("layers_6_linear_attn_in_proj_z_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1392115904)))]; + tensor linear_48_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = layers_6_linear_attn_in_proj_z_weight_to_fp16, x = x_155_cast_fp16)[name = string("linear_48_cast_fp16")]; + tensor var_4054 = const()[name = string("op_4054"), val = tensor([1, 1024, 32, 128])]; + tensor z_cast_fp16 = reshape(shape = var_4054, x = linear_48_cast_fp16)[name = string("z_cast_fp16")]; + tensor layers_6_linear_attn_in_proj_b_weight_to_fp16 = const()[name = string("layers_6_linear_attn_in_proj_b_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413087488)))]; + tensor linear_49_bias_0_to_fp16 = const()[name = string("linear_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413251392)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_49_bias_0_to_fp16, weight = layers_6_linear_attn_in_proj_b_weight_to_fp16, x = x_155_cast_fp16)[name = string("linear_49_cast_fp16")]; + tensor beta_21_cast_fp16 = sigmoid(x = linear_49_cast_fp16)[name = string("beta_21_cast_fp16")]; + tensor var_4063_weight_0_to_fp16 = const()[name = string("op_4063_weight_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413251520)))]; + tensor var_4063_bias_0_to_fp16 = const()[name = string("op_4063_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413415424)))]; + tensor var_4063_cast_fp16 = linear(bias = var_4063_bias_0_to_fp16, weight = var_4063_weight_0_to_fp16, x = x_155_cast_fp16)[name = string("op_4063_cast_fp16")]; + tensor var_4064_cast_fp16 = softplus(x = var_4063_cast_fp16)[name = string("op_4064_cast_fp16")]; + tensor var_4060_to_fp16 = const()[name = string("op_4060_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413415552)))]; + tensor g_21_cast_fp16 = mul(x = var_4060_to_fp16, y = var_4064_cast_fp16)[name = string("g_21_cast_fp16")]; + tensor var_4066_cast_fp16 = mul(x = q_55_cast_fp16, y = q_55_cast_fp16)[name = string("op_4066_cast_fp16")]; + tensor var_4068_axes_0 = const()[name = string("op_4068_axes_0"), val = tensor([-1])]; + bool var_4068_keep_dims_0 = const()[name = string("op_4068_keep_dims_0"), val = bool(true)]; + tensor var_4068_cast_fp16 = reduce_sum(axes = var_4068_axes_0, keep_dims = var_4068_keep_dims_0, x = var_4066_cast_fp16)[name = string("op_4068_cast_fp16")]; + fp16 var_4069_to_fp16 = const()[name = string("op_4069_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4070_cast_fp16 = add(x = var_4068_cast_fp16, y = var_4069_to_fp16)[name = string("op_4070_cast_fp16")]; + fp32 var_4071_epsilon_0 = const()[name = string("op_4071_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4071_cast_fp16 = rsqrt(epsilon = var_4071_epsilon_0, x = var_4070_cast_fp16)[name = string("op_4071_cast_fp16")]; + tensor q_57_cast_fp16 = mul(x = q_55_cast_fp16, y = var_4071_cast_fp16)[name = string("q_57_cast_fp16")]; + tensor var_4073_cast_fp16 = mul(x = k_47_cast_fp16, y = k_47_cast_fp16)[name = string("op_4073_cast_fp16")]; + tensor var_4075_axes_0 = const()[name = string("op_4075_axes_0"), val = tensor([-1])]; + bool var_4075_keep_dims_0 = const()[name = string("op_4075_keep_dims_0"), val = bool(true)]; + tensor var_4075_cast_fp16 = reduce_sum(axes = var_4075_axes_0, keep_dims = var_4075_keep_dims_0, x = var_4073_cast_fp16)[name = string("op_4075_cast_fp16")]; + fp16 var_4076_to_fp16 = const()[name = string("op_4076_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4077_cast_fp16 = add(x = var_4075_cast_fp16, y = var_4076_to_fp16)[name = string("op_4077_cast_fp16")]; + fp32 var_4078_epsilon_0 = const()[name = string("op_4078_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4078_cast_fp16 = rsqrt(epsilon = var_4078_epsilon_0, x = var_4077_cast_fp16)[name = string("op_4078_cast_fp16")]; + tensor k_49_cast_fp16 = mul(x = k_47_cast_fp16, y = var_4078_cast_fp16)[name = string("k_49_cast_fp16")]; + tensor var_4083_axes_0 = const()[name = string("op_4083_axes_0"), val = tensor([3])]; + tensor var_4083_cast_fp16 = expand_dims(axes = var_4083_axes_0, x = q_57_cast_fp16)[name = string("op_4083_cast_fp16")]; + tensor var_4085_reps_0 = const()[name = string("op_4085_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_4085_cast_fp16 = tile(reps = var_4085_reps_0, x = var_4083_cast_fp16)[name = string("op_4085_cast_fp16")]; + tensor var_4086 = const()[name = string("op_4086"), val = tensor([1, 1024, 32, 128])]; + tensor q_59_cast_fp16 = reshape(shape = var_4086, x = var_4085_cast_fp16)[name = string("q_59_cast_fp16")]; + tensor var_4091_axes_0 = const()[name = string("op_4091_axes_0"), val = tensor([3])]; + tensor var_4091_cast_fp16 = expand_dims(axes = var_4091_axes_0, x = k_49_cast_fp16)[name = string("op_4091_cast_fp16")]; + tensor var_4093_reps_0 = const()[name = string("op_4093_reps_0"), val = tensor([1, 1, 1, 2, 1])]; + tensor var_4093_cast_fp16 = tile(reps = var_4093_reps_0, x = var_4091_cast_fp16)[name = string("op_4093_cast_fp16")]; + tensor var_4094 = const()[name = string("op_4094"), val = tensor([1, 1024, 32, 128])]; + tensor k_51_cast_fp16 = reshape(shape = var_4094, x = var_4093_cast_fp16)[name = string("k_51_cast_fp16")]; + fp16 var_4096_to_fp16 = const()[name = string("op_4096_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor q_61_cast_fp16 = mul(x = q_59_cast_fp16, y = var_4096_to_fp16)[name = string("q_61_cast_fp16")]; + tensor var_4098_begin_0 = const()[name = string("op_4098_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4098_end_0 = const()[name = string("op_4098_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_4098_end_mask_0 = const()[name = string("op_4098_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_4098_squeeze_mask_0 = const()[name = string("op_4098_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_4098_cast_fp16 = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, squeeze_mask = var_4098_squeeze_mask_0, x = q_61_cast_fp16)[name = string("op_4098_cast_fp16")]; + tensor var_4099 = const()[name = string("op_4099"), val = tensor([1, 0, 2])]; + tensor var_4101 = const()[name = string("op_4101"), val = tensor([32, 16, 64, 128])]; + tensor var_4100_cast_fp16 = transpose(perm = var_4099, x = var_4098_cast_fp16)[name = string("transpose_9")]; + tensor q_63_cast_fp16 = reshape(shape = var_4101, x = var_4100_cast_fp16)[name = string("q_63_cast_fp16")]; + tensor var_4103_begin_0 = const()[name = string("op_4103_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4103_end_0 = const()[name = string("op_4103_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_4103_end_mask_0 = const()[name = string("op_4103_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_4103_squeeze_mask_0 = const()[name = string("op_4103_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_4103_cast_fp16 = slice_by_index(begin = var_4103_begin_0, end = var_4103_end_0, end_mask = var_4103_end_mask_0, squeeze_mask = var_4103_squeeze_mask_0, x = k_51_cast_fp16)[name = string("op_4103_cast_fp16")]; + tensor var_4104 = const()[name = string("op_4104"), val = tensor([1, 0, 2])]; + tensor var_4106 = const()[name = string("op_4106"), val = tensor([32, 16, 64, 128])]; + tensor var_4105_cast_fp16 = transpose(perm = var_4104, x = var_4103_cast_fp16)[name = string("transpose_8")]; + tensor k_53_cast_fp16 = reshape(shape = var_4106, x = var_4105_cast_fp16)[name = string("k_53_cast_fp16")]; + tensor var_4108_begin_0 = const()[name = string("op_4108_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4108_end_0 = const()[name = string("op_4108_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor var_4108_end_mask_0 = const()[name = string("op_4108_end_mask_0"), val = tensor([false, true, true, true])]; + tensor var_4108_squeeze_mask_0 = const()[name = string("op_4108_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor var_4108_cast_fp16 = slice_by_index(begin = var_4108_begin_0, end = var_4108_end_0, end_mask = var_4108_end_mask_0, squeeze_mask = var_4108_squeeze_mask_0, x = v_25_cast_fp16)[name = string("op_4108_cast_fp16")]; + tensor var_4109 = const()[name = string("op_4109"), val = tensor([1, 0, 2])]; + tensor var_4111 = const()[name = string("op_4111"), val = tensor([32, 16, 64, 128])]; + tensor var_4110_cast_fp16 = transpose(perm = var_4109, x = var_4108_cast_fp16)[name = string("transpose_7")]; + tensor v_27_cast_fp16 = reshape(shape = var_4111, x = var_4110_cast_fp16)[name = string("v_27_cast_fp16")]; + tensor var_4113_begin_0 = const()[name = string("op_4113_begin_0"), val = tensor([0, 0, 0])]; + tensor var_4113_end_0 = const()[name = string("op_4113_end_0"), val = tensor([1, 1024, 32])]; + tensor var_4113_end_mask_0 = const()[name = string("op_4113_end_mask_0"), val = tensor([false, true, true])]; + tensor var_4113_squeeze_mask_0 = const()[name = string("op_4113_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_4113_cast_fp16 = slice_by_index(begin = var_4113_begin_0, end = var_4113_end_0, end_mask = var_4113_end_mask_0, squeeze_mask = var_4113_squeeze_mask_0, x = beta_21_cast_fp16)[name = string("op_4113_cast_fp16")]; + tensor var_4114_perm_0 = const()[name = string("op_4114_perm_0"), val = tensor([1, 0])]; + tensor var_4115 = const()[name = string("op_4115"), val = tensor([32, 16, 64, 1])]; + tensor var_4114_cast_fp16 = transpose(perm = var_4114_perm_0, x = var_4113_cast_fp16)[name = string("transpose_6")]; + tensor beta_cast_fp16 = reshape(shape = var_4115, x = var_4114_cast_fp16)[name = string("beta_cast_fp16")]; + tensor var_4117_begin_0 = const()[name = string("op_4117_begin_0"), val = tensor([0, 0, 0])]; + tensor var_4117_end_0 = const()[name = string("op_4117_end_0"), val = tensor([1, 1024, 32])]; + tensor var_4117_end_mask_0 = const()[name = string("op_4117_end_mask_0"), val = tensor([false, true, true])]; + tensor var_4117_squeeze_mask_0 = const()[name = string("op_4117_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_4117_cast_fp16 = slice_by_index(begin = var_4117_begin_0, end = var_4117_end_0, end_mask = var_4117_end_mask_0, squeeze_mask = var_4117_squeeze_mask_0, x = g_21_cast_fp16)[name = string("op_4117_cast_fp16")]; + tensor var_4118_perm_0 = const()[name = string("op_4118_perm_0"), val = tensor([1, 0])]; + tensor var_4119 = const()[name = string("op_4119"), val = tensor([32, 16, 64])]; + tensor var_4118_cast_fp16 = transpose(perm = var_4118_perm_0, x = var_4117_cast_fp16)[name = string("transpose_5")]; + tensor g_cast_fp16 = reshape(shape = var_4119, x = var_4118_cast_fp16)[name = string("g_cast_fp16")]; + tensor g_col_axes_0 = const()[name = string("g_col_axes_0"), val = tensor([-1])]; + tensor g_col_cast_fp16 = expand_dims(axes = g_col_axes_0, x = g_cast_fp16)[name = string("g_col_cast_fp16")]; + bool var_4122_transpose_x_0 = const()[name = string("op_4122_transpose_x_0"), val = bool(false)]; + bool var_4122_transpose_y_0 = const()[name = string("op_4122_transpose_y_0"), val = bool(false)]; + tensor var_4122_cast_fp16 = matmul(transpose_x = var_4122_transpose_x_0, transpose_y = var_4122_transpose_y_0, x = layers_0_linear_attn_lower_incl_to_fp16, y = g_col_cast_fp16)[name = string("op_4122_cast_fp16")]; + tensor cum_axes_0 = const()[name = string("cum_axes_0"), val = tensor([-1])]; + tensor cum_cast_fp16 = squeeze(axes = cum_axes_0, x = var_4122_cast_fp16)[name = string("cum_cast_fp16")]; + bool var_4124_transpose_x_0 = const()[name = string("op_4124_transpose_x_0"), val = bool(false)]; + bool var_4124_transpose_y_0 = const()[name = string("op_4124_transpose_y_0"), val = bool(false)]; + tensor var_4124_cast_fp16 = matmul(transpose_x = var_4124_transpose_x_0, transpose_y = var_4124_transpose_y_0, x = layers_0_linear_attn_suffix_sum_t_to_fp16, y = g_col_cast_fp16)[name = string("op_4124_cast_fp16")]; + tensor to_end_axes_0 = const()[name = string("to_end_axes_0"), val = tensor([-1])]; + tensor to_end_cast_fp16 = squeeze(axes = to_end_axes_0, x = var_4124_cast_fp16)[name = string("to_end_cast_fp16")]; + bool var_4126_transpose_x_0 = const()[name = string("op_4126_transpose_x_0"), val = bool(false)]; + bool var_4126_transpose_y_0 = const()[name = string("op_4126_transpose_y_0"), val = bool(false)]; + tensor var_4126_cast_fp16 = matmul(transpose_x = var_4126_transpose_x_0, transpose_y = var_4126_transpose_y_0, x = layers_0_linear_attn_pair_sum_t_to_fp16, y = g_col_cast_fp16)[name = string("op_4126_cast_fp16")]; + tensor var_4127 = const()[name = string("op_4127"), val = tensor([32, 16, 64, 64])]; + tensor pair_71_cast_fp16 = reshape(shape = var_4127, x = var_4126_cast_fp16)[name = string("pair_71_cast_fp16")]; + tensor var_4129_cast_fp16 = exp(x = pair_71_cast_fp16)[name = string("op_4129_cast_fp16")]; + tensor pair_decay_cast_fp16 = mul(x = var_4129_cast_fp16, y = layers_0_linear_attn_lower_incl_to_fp16)[name = string("pair_decay_cast_fp16")]; + tensor k_beta_cast_fp16 = mul(x = k_53_cast_fp16, y = beta_cast_fp16)[name = string("k_beta_cast_fp16")]; + tensor v_beta_cast_fp16 = mul(x = v_27_cast_fp16, y = beta_cast_fp16)[name = string("v_beta_cast_fp16")]; + bool var_4134_transpose_x_1 = const()[name = string("op_4134_transpose_x_1"), val = bool(false)]; + bool var_4134_transpose_y_1 = const()[name = string("op_4134_transpose_y_1"), val = bool(true)]; + tensor var_4134_cast_fp16 = matmul(transpose_x = var_4134_transpose_x_1, transpose_y = var_4134_transpose_y_1, x = k_beta_cast_fp16, y = k_53_cast_fp16)[name = string("op_4134_cast_fp16")]; + tensor kkt_cast_fp16 = mul(x = var_4134_cast_fp16, y = pair_decay_cast_fp16)[name = string("kkt_cast_fp16")]; + bool var_4137_transpose_x_1 = const()[name = string("op_4137_transpose_x_1"), val = bool(false)]; + bool var_4137_transpose_y_1 = const()[name = string("op_4137_transpose_y_1"), val = bool(true)]; + tensor var_4137_cast_fp16 = matmul(transpose_x = var_4137_transpose_x_1, transpose_y = var_4137_transpose_y_1, x = q_63_cast_fp16, y = k_53_cast_fp16)[name = string("op_4137_cast_fp16")]; + tensor attn_intra_cast_fp16 = mul(x = var_4137_cast_fp16, y = pair_decay_cast_fp16)[name = string("attn_intra_cast_fp16")]; + tensor var_4139_cast_fp16 = mul(x = kkt_cast_fp16, y = layers_0_linear_attn_strict_lower_to_fp16)[name = string("op_4139_cast_fp16")]; + tensor t_21_cast_fp16 = add(x = layers_0_linear_attn_eye_to_fp16, y = var_4139_cast_fp16)[name = string("t_21_cast_fp16")]; + tensor var_4143 = const()[name = string("op_4143"), val = tensor([-1, 64, 64])]; + tensor t_cast_fp16 = reshape(shape = var_4143, x = t_21_cast_fp16)[name = string("t_cast_fp16")]; + tensor var_4148 = const()[name = string("op_4148"), val = tensor([512, 32, 2, 32, 2])]; + tensor var_4149_cast_fp16 = reshape(shape = var_4148, x = t_cast_fp16)[name = string("op_4149_cast_fp16")]; + tensor var_4154_cast_fp16 = mul(x = var_4149_cast_fp16, y = var_216_to_fp16)[name = string("op_4154_cast_fp16")]; + tensor blocks_61_axes_0 = const()[name = string("blocks_61_axes_0"), val = tensor([-2])]; + bool blocks_61_keep_dims_0 = const()[name = string("blocks_61_keep_dims_0"), val = bool(false)]; + tensor blocks_61_cast_fp16 = reduce_sum(axes = blocks_61_axes_0, keep_dims = blocks_61_keep_dims_0, x = var_4154_cast_fp16)[name = string("blocks_61_cast_fp16")]; + tensor var_4157_begin_0 = const()[name = string("op_4157_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4157_end_0 = const()[name = string("op_4157_end_0"), val = tensor([512, 32, 2, 2])]; + tensor var_4157_end_mask_0 = const()[name = string("op_4157_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4157_cast_fp16 = slice_by_index(begin = var_4157_begin_0, end = var_4157_end_0, end_mask = var_4157_end_mask_0, x = blocks_61_cast_fp16)[name = string("op_4157_cast_fp16")]; + tensor x_157_begin_0 = const()[name = string("x_157_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_157_end_0 = const()[name = string("x_157_end_0"), val = tensor([512, 32, 1, 1])]; + tensor x_157_end_mask_0 = const()[name = string("x_157_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_157_cast_fp16 = slice_by_index(begin = x_157_begin_0, end = x_157_end_0, end_mask = x_157_end_mask_0, x = var_4157_cast_fp16)[name = string("x_157_cast_fp16")]; + bool var_4167_transpose_x_0 = const()[name = string("op_4167_transpose_x_0"), val = bool(false)]; + bool var_4167_transpose_y_0 = const()[name = string("op_4167_transpose_y_0"), val = bool(false)]; + tensor var_4167_cast_fp16 = matmul(transpose_x = var_4167_transpose_x_0, transpose_y = var_4167_transpose_y_0, x = x_157_cast_fp16, y = var_224_to_fp16)[name = string("op_4167_cast_fp16")]; + bool var_4168_transpose_x_0 = const()[name = string("op_4168_transpose_x_0"), val = bool(false)]; + bool var_4168_transpose_y_0 = const()[name = string("op_4168_transpose_y_0"), val = bool(false)]; + tensor var_4168_cast_fp16 = matmul(transpose_x = var_4168_transpose_x_0, transpose_y = var_4168_transpose_y_0, x = var_224_to_fp16, y = var_4167_cast_fp16)[name = string("op_4168_cast_fp16")]; + fp16 const_71_promoted_to_fp16 = const()[name = string("const_71_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_61_cast_fp16 = mul(x = var_4168_cast_fp16, y = const_71_promoted_to_fp16)[name = string("lower_61_cast_fp16")]; + bool bottom_61_interleave_0 = const()[name = string("bottom_61_interleave_0"), val = bool(false)]; + tensor bottom_61_cast_fp16 = concat(axis = var_3995, interleave = bottom_61_interleave_0, values = (lower_61_cast_fp16, var_224_to_fp16))[name = string("bottom_61_cast_fp16")]; + bool inv_83_interleave_0 = const()[name = string("inv_83_interleave_0"), val = bool(false)]; + tensor inv_83_cast_fp16 = concat(axis = var_3974, interleave = inv_83_interleave_0, values = (top_1_to_fp16, bottom_61_cast_fp16))[name = string("inv_83_cast_fp16")]; + tensor var_4177 = const()[name = string("op_4177"), val = tensor([512, 16, 4, 16, 4])]; + tensor var_4178_cast_fp16 = reshape(shape = var_4177, x = t_cast_fp16)[name = string("op_4178_cast_fp16")]; + tensor var_4183_cast_fp16 = mul(x = var_4178_cast_fp16, y = var_245_to_fp16)[name = string("op_4183_cast_fp16")]; + tensor blocks_63_axes_0 = const()[name = string("blocks_63_axes_0"), val = tensor([-2])]; + bool blocks_63_keep_dims_0 = const()[name = string("blocks_63_keep_dims_0"), val = bool(false)]; + tensor blocks_63_cast_fp16 = reduce_sum(axes = blocks_63_axes_0, keep_dims = blocks_63_keep_dims_0, x = var_4183_cast_fp16)[name = string("blocks_63_cast_fp16")]; + tensor var_4186_begin_0 = const()[name = string("op_4186_begin_0"), val = tensor([0, 0, 2, 0])]; + tensor var_4186_end_0 = const()[name = string("op_4186_end_0"), val = tensor([512, 16, 4, 4])]; + tensor var_4186_end_mask_0 = const()[name = string("op_4186_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4186_cast_fp16 = slice_by_index(begin = var_4186_begin_0, end = var_4186_end_0, end_mask = var_4186_end_mask_0, x = blocks_63_cast_fp16)[name = string("op_4186_cast_fp16")]; + tensor x_159_begin_0 = const()[name = string("x_159_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_159_end_0 = const()[name = string("x_159_end_0"), val = tensor([512, 16, 2, 2])]; + tensor x_159_end_mask_0 = const()[name = string("x_159_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_159_cast_fp16 = slice_by_index(begin = x_159_begin_0, end = x_159_end_0, end_mask = x_159_end_mask_0, x = var_4186_cast_fp16)[name = string("x_159_cast_fp16")]; + tensor var_4188 = const()[name = string("op_4188"), val = tensor([512, 16, 2, 2, 2])]; + tensor pair_75_cast_fp16 = reshape(shape = var_4188, x = inv_83_cast_fp16)[name = string("pair_75_cast_fp16")]; + tensor var_4190_begin_0 = const()[name = string("op_4190_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_4190_end_0 = const()[name = string("op_4190_end_0"), val = tensor([512, 16, 1, 2, 2])]; + tensor var_4190_end_mask_0 = const()[name = string("op_4190_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4190_squeeze_mask_0 = const()[name = string("op_4190_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4190_cast_fp16 = slice_by_index(begin = var_4190_begin_0, end = var_4190_end_0, end_mask = var_4190_end_mask_0, squeeze_mask = var_4190_squeeze_mask_0, x = pair_75_cast_fp16)[name = string("op_4190_cast_fp16")]; + tensor var_4193_begin_0 = const()[name = string("op_4193_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_4193_end_0 = const()[name = string("op_4193_end_0"), val = tensor([512, 16, 2, 2, 2])]; + tensor var_4193_end_mask_0 = const()[name = string("op_4193_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4193_squeeze_mask_0 = const()[name = string("op_4193_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4193_cast_fp16 = slice_by_index(begin = var_4193_begin_0, end = var_4193_end_0, end_mask = var_4193_end_mask_0, squeeze_mask = var_4193_squeeze_mask_0, x = pair_75_cast_fp16)[name = string("op_4193_cast_fp16")]; + bool var_4196_transpose_x_0 = const()[name = string("op_4196_transpose_x_0"), val = bool(false)]; + bool var_4196_transpose_y_0 = const()[name = string("op_4196_transpose_y_0"), val = bool(false)]; + tensor var_4196_cast_fp16 = matmul(transpose_x = var_4196_transpose_x_0, transpose_y = var_4196_transpose_y_0, x = x_159_cast_fp16, y = var_4190_cast_fp16)[name = string("op_4196_cast_fp16")]; + bool var_4197_transpose_x_0 = const()[name = string("op_4197_transpose_x_0"), val = bool(false)]; + bool var_4197_transpose_y_0 = const()[name = string("op_4197_transpose_y_0"), val = bool(false)]; + tensor var_4197_cast_fp16 = matmul(transpose_x = var_4197_transpose_x_0, transpose_y = var_4197_transpose_y_0, x = var_4193_cast_fp16, y = var_4196_cast_fp16)[name = string("op_4197_cast_fp16")]; + fp16 const_72_promoted_to_fp16 = const()[name = string("const_72_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_63_cast_fp16 = mul(x = var_4197_cast_fp16, y = const_72_promoted_to_fp16)[name = string("lower_63_cast_fp16")]; + bool top_63_interleave_0 = const()[name = string("top_63_interleave_0"), val = bool(false)]; + tensor top_63_cast_fp16 = concat(axis = var_3995, interleave = top_63_interleave_0, values = (var_4190_cast_fp16, var_262_to_fp16))[name = string("top_63_cast_fp16")]; + bool bottom_63_interleave_0 = const()[name = string("bottom_63_interleave_0"), val = bool(false)]; + tensor bottom_63_cast_fp16 = concat(axis = var_3995, interleave = bottom_63_interleave_0, values = (lower_63_cast_fp16, var_4193_cast_fp16))[name = string("bottom_63_cast_fp16")]; + bool inv_85_interleave_0 = const()[name = string("inv_85_interleave_0"), val = bool(false)]; + tensor inv_85_cast_fp16 = concat(axis = var_3974, interleave = inv_85_interleave_0, values = (top_63_cast_fp16, bottom_63_cast_fp16))[name = string("inv_85_cast_fp16")]; + tensor var_4206 = const()[name = string("op_4206"), val = tensor([512, 8, 8, 8, 8])]; + tensor var_4207_cast_fp16 = reshape(shape = var_4206, x = t_cast_fp16)[name = string("op_4207_cast_fp16")]; + tensor var_4211_to_fp16 = const()[name = string("op_4211_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413415680)))]; + tensor var_4212_cast_fp16 = mul(x = var_4207_cast_fp16, y = var_4211_to_fp16)[name = string("op_4212_cast_fp16")]; + tensor blocks_65_axes_0 = const()[name = string("blocks_65_axes_0"), val = tensor([-2])]; + bool blocks_65_keep_dims_0 = const()[name = string("blocks_65_keep_dims_0"), val = bool(false)]; + tensor blocks_65_cast_fp16 = reduce_sum(axes = blocks_65_axes_0, keep_dims = blocks_65_keep_dims_0, x = var_4212_cast_fp16)[name = string("blocks_65_cast_fp16")]; + tensor var_4215_begin_0 = const()[name = string("op_4215_begin_0"), val = tensor([0, 0, 4, 0])]; + tensor var_4215_end_0 = const()[name = string("op_4215_end_0"), val = tensor([512, 8, 8, 8])]; + tensor var_4215_end_mask_0 = const()[name = string("op_4215_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4215_cast_fp16 = slice_by_index(begin = var_4215_begin_0, end = var_4215_end_0, end_mask = var_4215_end_mask_0, x = blocks_65_cast_fp16)[name = string("op_4215_cast_fp16")]; + tensor x_161_begin_0 = const()[name = string("x_161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_161_end_0 = const()[name = string("x_161_end_0"), val = tensor([512, 8, 4, 4])]; + tensor x_161_end_mask_0 = const()[name = string("x_161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_161_cast_fp16 = slice_by_index(begin = x_161_begin_0, end = x_161_end_0, end_mask = x_161_end_mask_0, x = var_4215_cast_fp16)[name = string("x_161_cast_fp16")]; + tensor var_4217 = const()[name = string("op_4217"), val = tensor([512, 8, 2, 4, 4])]; + tensor pair_77_cast_fp16 = reshape(shape = var_4217, x = inv_85_cast_fp16)[name = string("pair_77_cast_fp16")]; + tensor var_4219_begin_0 = const()[name = string("op_4219_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_4219_end_0 = const()[name = string("op_4219_end_0"), val = tensor([512, 8, 1, 4, 4])]; + tensor var_4219_end_mask_0 = const()[name = string("op_4219_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4219_squeeze_mask_0 = const()[name = string("op_4219_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4219_cast_fp16 = slice_by_index(begin = var_4219_begin_0, end = var_4219_end_0, end_mask = var_4219_end_mask_0, squeeze_mask = var_4219_squeeze_mask_0, x = pair_77_cast_fp16)[name = string("op_4219_cast_fp16")]; + tensor var_4222_begin_0 = const()[name = string("op_4222_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_4222_end_0 = const()[name = string("op_4222_end_0"), val = tensor([512, 8, 2, 4, 4])]; + tensor var_4222_end_mask_0 = const()[name = string("op_4222_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4222_squeeze_mask_0 = const()[name = string("op_4222_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4222_cast_fp16 = slice_by_index(begin = var_4222_begin_0, end = var_4222_end_0, end_mask = var_4222_end_mask_0, squeeze_mask = var_4222_squeeze_mask_0, x = pair_77_cast_fp16)[name = string("op_4222_cast_fp16")]; + bool var_4225_transpose_x_0 = const()[name = string("op_4225_transpose_x_0"), val = bool(false)]; + bool var_4225_transpose_y_0 = const()[name = string("op_4225_transpose_y_0"), val = bool(false)]; + tensor var_4225_cast_fp16 = matmul(transpose_x = var_4225_transpose_x_0, transpose_y = var_4225_transpose_y_0, x = x_161_cast_fp16, y = var_4219_cast_fp16)[name = string("op_4225_cast_fp16")]; + bool var_4226_transpose_x_0 = const()[name = string("op_4226_transpose_x_0"), val = bool(false)]; + bool var_4226_transpose_y_0 = const()[name = string("op_4226_transpose_y_0"), val = bool(false)]; + tensor var_4226_cast_fp16 = matmul(transpose_x = var_4226_transpose_x_0, transpose_y = var_4226_transpose_y_0, x = var_4222_cast_fp16, y = var_4225_cast_fp16)[name = string("op_4226_cast_fp16")]; + fp16 const_73_promoted_to_fp16 = const()[name = string("const_73_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_65_cast_fp16 = mul(x = var_4226_cast_fp16, y = const_73_promoted_to_fp16)[name = string("lower_65_cast_fp16")]; + bool top_65_interleave_0 = const()[name = string("top_65_interleave_0"), val = bool(false)]; + tensor top_65_cast_fp16 = concat(axis = var_3995, interleave = top_65_interleave_0, values = (var_4219_cast_fp16, var_291_to_fp16))[name = string("top_65_cast_fp16")]; + bool bottom_65_interleave_0 = const()[name = string("bottom_65_interleave_0"), val = bool(false)]; + tensor bottom_65_cast_fp16 = concat(axis = var_3995, interleave = bottom_65_interleave_0, values = (lower_65_cast_fp16, var_4222_cast_fp16))[name = string("bottom_65_cast_fp16")]; + bool inv_87_interleave_0 = const()[name = string("inv_87_interleave_0"), val = bool(false)]; + tensor inv_87_cast_fp16 = concat(axis = var_3974, interleave = inv_87_interleave_0, values = (top_65_cast_fp16, bottom_65_cast_fp16))[name = string("inv_87_cast_fp16")]; + tensor var_4235 = const()[name = string("op_4235"), val = tensor([512, 4, 16, 4, 16])]; + tensor var_4236_cast_fp16 = reshape(shape = var_4235, x = t_cast_fp16)[name = string("op_4236_cast_fp16")]; + tensor var_4240_to_fp16 = const()[name = string("op_4240_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413415872)))]; + tensor var_4241_cast_fp16 = mul(x = var_4236_cast_fp16, y = var_4240_to_fp16)[name = string("op_4241_cast_fp16")]; + tensor blocks_67_axes_0 = const()[name = string("blocks_67_axes_0"), val = tensor([-2])]; + bool blocks_67_keep_dims_0 = const()[name = string("blocks_67_keep_dims_0"), val = bool(false)]; + tensor blocks_67_cast_fp16 = reduce_sum(axes = blocks_67_axes_0, keep_dims = blocks_67_keep_dims_0, x = var_4241_cast_fp16)[name = string("blocks_67_cast_fp16")]; + tensor var_4244_begin_0 = const()[name = string("op_4244_begin_0"), val = tensor([0, 0, 8, 0])]; + tensor var_4244_end_0 = const()[name = string("op_4244_end_0"), val = tensor([512, 4, 16, 16])]; + tensor var_4244_end_mask_0 = const()[name = string("op_4244_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4244_cast_fp16 = slice_by_index(begin = var_4244_begin_0, end = var_4244_end_0, end_mask = var_4244_end_mask_0, x = blocks_67_cast_fp16)[name = string("op_4244_cast_fp16")]; + tensor x_163_begin_0 = const()[name = string("x_163_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_163_end_0 = const()[name = string("x_163_end_0"), val = tensor([512, 4, 8, 8])]; + tensor x_163_end_mask_0 = const()[name = string("x_163_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_163_cast_fp16 = slice_by_index(begin = x_163_begin_0, end = x_163_end_0, end_mask = x_163_end_mask_0, x = var_4244_cast_fp16)[name = string("x_163_cast_fp16")]; + tensor var_4246 = const()[name = string("op_4246"), val = tensor([512, 4, 2, 8, 8])]; + tensor pair_79_cast_fp16 = reshape(shape = var_4246, x = inv_87_cast_fp16)[name = string("pair_79_cast_fp16")]; + tensor var_4248_begin_0 = const()[name = string("op_4248_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_4248_end_0 = const()[name = string("op_4248_end_0"), val = tensor([512, 4, 1, 8, 8])]; + tensor var_4248_end_mask_0 = const()[name = string("op_4248_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4248_squeeze_mask_0 = const()[name = string("op_4248_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4248_cast_fp16 = slice_by_index(begin = var_4248_begin_0, end = var_4248_end_0, end_mask = var_4248_end_mask_0, squeeze_mask = var_4248_squeeze_mask_0, x = pair_79_cast_fp16)[name = string("op_4248_cast_fp16")]; + tensor var_4251_begin_0 = const()[name = string("op_4251_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_4251_end_0 = const()[name = string("op_4251_end_0"), val = tensor([512, 4, 2, 8, 8])]; + tensor var_4251_end_mask_0 = const()[name = string("op_4251_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4251_squeeze_mask_0 = const()[name = string("op_4251_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4251_cast_fp16 = slice_by_index(begin = var_4251_begin_0, end = var_4251_end_0, end_mask = var_4251_end_mask_0, squeeze_mask = var_4251_squeeze_mask_0, x = pair_79_cast_fp16)[name = string("op_4251_cast_fp16")]; + bool var_4254_transpose_x_0 = const()[name = string("op_4254_transpose_x_0"), val = bool(false)]; + bool var_4254_transpose_y_0 = const()[name = string("op_4254_transpose_y_0"), val = bool(false)]; + tensor var_4254_cast_fp16 = matmul(transpose_x = var_4254_transpose_x_0, transpose_y = var_4254_transpose_y_0, x = x_163_cast_fp16, y = var_4248_cast_fp16)[name = string("op_4254_cast_fp16")]; + bool var_4255_transpose_x_0 = const()[name = string("op_4255_transpose_x_0"), val = bool(false)]; + bool var_4255_transpose_y_0 = const()[name = string("op_4255_transpose_y_0"), val = bool(false)]; + tensor var_4255_cast_fp16 = matmul(transpose_x = var_4255_transpose_x_0, transpose_y = var_4255_transpose_y_0, x = var_4251_cast_fp16, y = var_4254_cast_fp16)[name = string("op_4255_cast_fp16")]; + fp16 const_74_promoted_to_fp16 = const()[name = string("const_74_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_67_cast_fp16 = mul(x = var_4255_cast_fp16, y = const_74_promoted_to_fp16)[name = string("lower_67_cast_fp16")]; + bool top_67_interleave_0 = const()[name = string("top_67_interleave_0"), val = bool(false)]; + tensor top_67_cast_fp16 = concat(axis = var_3995, interleave = top_67_interleave_0, values = (var_4248_cast_fp16, var_320_to_fp16))[name = string("top_67_cast_fp16")]; + bool bottom_67_interleave_0 = const()[name = string("bottom_67_interleave_0"), val = bool(false)]; + tensor bottom_67_cast_fp16 = concat(axis = var_3995, interleave = bottom_67_interleave_0, values = (lower_67_cast_fp16, var_4251_cast_fp16))[name = string("bottom_67_cast_fp16")]; + bool inv_89_interleave_0 = const()[name = string("inv_89_interleave_0"), val = bool(false)]; + tensor inv_89_cast_fp16 = concat(axis = var_3974, interleave = inv_89_interleave_0, values = (top_67_cast_fp16, bottom_67_cast_fp16))[name = string("inv_89_cast_fp16")]; + tensor var_4264 = const()[name = string("op_4264"), val = tensor([512, 2, 32, 2, 32])]; + tensor var_4265_cast_fp16 = reshape(shape = var_4264, x = t_cast_fp16)[name = string("op_4265_cast_fp16")]; + tensor var_4269_to_fp16 = const()[name = string("op_4269_to_fp16"), val = tensor([[[[0x1p+0], [0x0p+0]]], [[[0x0p+0], [0x1p+0]]]])]; + tensor var_4270_cast_fp16 = mul(x = var_4265_cast_fp16, y = var_4269_to_fp16)[name = string("op_4270_cast_fp16")]; + tensor blocks_69_axes_0 = const()[name = string("blocks_69_axes_0"), val = tensor([-2])]; + bool blocks_69_keep_dims_0 = const()[name = string("blocks_69_keep_dims_0"), val = bool(false)]; + tensor blocks_69_cast_fp16 = reduce_sum(axes = blocks_69_axes_0, keep_dims = blocks_69_keep_dims_0, x = var_4270_cast_fp16)[name = string("blocks_69_cast_fp16")]; + tensor var_4273_begin_0 = const()[name = string("op_4273_begin_0"), val = tensor([0, 0, 16, 0])]; + tensor var_4273_end_0 = const()[name = string("op_4273_end_0"), val = tensor([512, 2, 32, 32])]; + tensor var_4273_end_mask_0 = const()[name = string("op_4273_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4273_cast_fp16 = slice_by_index(begin = var_4273_begin_0, end = var_4273_end_0, end_mask = var_4273_end_mask_0, x = blocks_69_cast_fp16)[name = string("op_4273_cast_fp16")]; + tensor x_165_begin_0 = const()[name = string("x_165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_165_end_0 = const()[name = string("x_165_end_0"), val = tensor([512, 2, 16, 16])]; + tensor x_165_end_mask_0 = const()[name = string("x_165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_165_cast_fp16 = slice_by_index(begin = x_165_begin_0, end = x_165_end_0, end_mask = x_165_end_mask_0, x = var_4273_cast_fp16)[name = string("x_165_cast_fp16")]; + tensor var_4275 = const()[name = string("op_4275"), val = tensor([512, 2, 2, 16, 16])]; + tensor pair_81_cast_fp16 = reshape(shape = var_4275, x = inv_89_cast_fp16)[name = string("pair_81_cast_fp16")]; + tensor var_4277_begin_0 = const()[name = string("op_4277_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_4277_end_0 = const()[name = string("op_4277_end_0"), val = tensor([512, 2, 1, 16, 16])]; + tensor var_4277_end_mask_0 = const()[name = string("op_4277_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4277_squeeze_mask_0 = const()[name = string("op_4277_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4277_cast_fp16 = slice_by_index(begin = var_4277_begin_0, end = var_4277_end_0, end_mask = var_4277_end_mask_0, squeeze_mask = var_4277_squeeze_mask_0, x = pair_81_cast_fp16)[name = string("op_4277_cast_fp16")]; + tensor var_4280_begin_0 = const()[name = string("op_4280_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_4280_end_0 = const()[name = string("op_4280_end_0"), val = tensor([512, 2, 2, 16, 16])]; + tensor var_4280_end_mask_0 = const()[name = string("op_4280_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4280_squeeze_mask_0 = const()[name = string("op_4280_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4280_cast_fp16 = slice_by_index(begin = var_4280_begin_0, end = var_4280_end_0, end_mask = var_4280_end_mask_0, squeeze_mask = var_4280_squeeze_mask_0, x = pair_81_cast_fp16)[name = string("op_4280_cast_fp16")]; + bool var_4283_transpose_x_0 = const()[name = string("op_4283_transpose_x_0"), val = bool(false)]; + bool var_4283_transpose_y_0 = const()[name = string("op_4283_transpose_y_0"), val = bool(false)]; + tensor var_4283_cast_fp16 = matmul(transpose_x = var_4283_transpose_x_0, transpose_y = var_4283_transpose_y_0, x = x_165_cast_fp16, y = var_4277_cast_fp16)[name = string("op_4283_cast_fp16")]; + bool var_4284_transpose_x_0 = const()[name = string("op_4284_transpose_x_0"), val = bool(false)]; + bool var_4284_transpose_y_0 = const()[name = string("op_4284_transpose_y_0"), val = bool(false)]; + tensor var_4284_cast_fp16 = matmul(transpose_x = var_4284_transpose_x_0, transpose_y = var_4284_transpose_y_0, x = var_4280_cast_fp16, y = var_4283_cast_fp16)[name = string("op_4284_cast_fp16")]; + fp16 const_75_promoted_to_fp16 = const()[name = string("const_75_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_69_cast_fp16 = mul(x = var_4284_cast_fp16, y = const_75_promoted_to_fp16)[name = string("lower_69_cast_fp16")]; + bool top_69_interleave_0 = const()[name = string("top_69_interleave_0"), val = bool(false)]; + tensor top_69_cast_fp16 = concat(axis = var_3995, interleave = top_69_interleave_0, values = (var_4277_cast_fp16, var_349_to_fp16))[name = string("top_69_cast_fp16")]; + bool bottom_69_interleave_0 = const()[name = string("bottom_69_interleave_0"), val = bool(false)]; + tensor bottom_69_cast_fp16 = concat(axis = var_3995, interleave = bottom_69_interleave_0, values = (lower_69_cast_fp16, var_4280_cast_fp16))[name = string("bottom_69_cast_fp16")]; + bool inv_91_interleave_0 = const()[name = string("inv_91_interleave_0"), val = bool(false)]; + tensor inv_91_cast_fp16 = concat(axis = var_3974, interleave = inv_91_interleave_0, values = (top_69_cast_fp16, bottom_69_cast_fp16))[name = string("inv_91_cast_fp16")]; + tensor var_4293 = const()[name = string("op_4293"), val = tensor([512, 1, 64, 1, 64])]; + tensor var_4294_cast_fp16 = reshape(shape = var_4293, x = t_cast_fp16)[name = string("op_4294_cast_fp16")]; + tensor blocks_axes_0 = const()[name = string("blocks_axes_0"), val = tensor([-2])]; + bool blocks_keep_dims_0 = const()[name = string("blocks_keep_dims_0"), val = bool(false)]; + tensor blocks_cast_fp16 = reduce_sum(axes = blocks_axes_0, keep_dims = blocks_keep_dims_0, x = var_4294_cast_fp16)[name = string("blocks_cast_fp16")]; + tensor var_4302_begin_0 = const()[name = string("op_4302_begin_0"), val = tensor([0, 0, 32, 0])]; + tensor var_4302_end_0 = const()[name = string("op_4302_end_0"), val = tensor([512, 1, 64, 64])]; + tensor var_4302_end_mask_0 = const()[name = string("op_4302_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4302_cast_fp16 = slice_by_index(begin = var_4302_begin_0, end = var_4302_end_0, end_mask = var_4302_end_mask_0, x = blocks_cast_fp16)[name = string("op_4302_cast_fp16")]; + tensor x_167_begin_0 = const()[name = string("x_167_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_167_end_0 = const()[name = string("x_167_end_0"), val = tensor([512, 1, 32, 32])]; + tensor x_167_end_mask_0 = const()[name = string("x_167_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_167_cast_fp16 = slice_by_index(begin = x_167_begin_0, end = x_167_end_0, end_mask = x_167_end_mask_0, x = var_4302_cast_fp16)[name = string("x_167_cast_fp16")]; + tensor var_4304 = const()[name = string("op_4304"), val = tensor([512, 1, 2, 32, 32])]; + tensor pair_cast_fp16 = reshape(shape = var_4304, x = inv_91_cast_fp16)[name = string("pair_cast_fp16")]; + tensor var_4306_begin_0 = const()[name = string("op_4306_begin_0"), val = tensor([0, 0, 0, 0, 0])]; + tensor var_4306_end_0 = const()[name = string("op_4306_end_0"), val = tensor([512, 1, 1, 32, 32])]; + tensor var_4306_end_mask_0 = const()[name = string("op_4306_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4306_squeeze_mask_0 = const()[name = string("op_4306_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4306_cast_fp16 = slice_by_index(begin = var_4306_begin_0, end = var_4306_end_0, end_mask = var_4306_end_mask_0, squeeze_mask = var_4306_squeeze_mask_0, x = pair_cast_fp16)[name = string("op_4306_cast_fp16")]; + tensor var_4309_begin_0 = const()[name = string("op_4309_begin_0"), val = tensor([0, 0, 1, 0, 0])]; + tensor var_4309_end_0 = const()[name = string("op_4309_end_0"), val = tensor([512, 1, 2, 32, 32])]; + tensor var_4309_end_mask_0 = const()[name = string("op_4309_end_mask_0"), val = tensor([true, true, false, true, true])]; + tensor var_4309_squeeze_mask_0 = const()[name = string("op_4309_squeeze_mask_0"), val = tensor([false, false, true, false, false])]; + tensor var_4309_cast_fp16 = slice_by_index(begin = var_4309_begin_0, end = var_4309_end_0, end_mask = var_4309_end_mask_0, squeeze_mask = var_4309_squeeze_mask_0, x = pair_cast_fp16)[name = string("op_4309_cast_fp16")]; + bool var_4312_transpose_x_0 = const()[name = string("op_4312_transpose_x_0"), val = bool(false)]; + bool var_4312_transpose_y_0 = const()[name = string("op_4312_transpose_y_0"), val = bool(false)]; + tensor var_4312_cast_fp16 = matmul(transpose_x = var_4312_transpose_x_0, transpose_y = var_4312_transpose_y_0, x = x_167_cast_fp16, y = var_4306_cast_fp16)[name = string("op_4312_cast_fp16")]; + bool var_4313_transpose_x_0 = const()[name = string("op_4313_transpose_x_0"), val = bool(false)]; + bool var_4313_transpose_y_0 = const()[name = string("op_4313_transpose_y_0"), val = bool(false)]; + tensor var_4313_cast_fp16 = matmul(transpose_x = var_4313_transpose_x_0, transpose_y = var_4313_transpose_y_0, x = var_4309_cast_fp16, y = var_4312_cast_fp16)[name = string("op_4313_cast_fp16")]; + fp16 const_76_promoted_to_fp16 = const()[name = string("const_76_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor lower_cast_fp16 = mul(x = var_4313_cast_fp16, y = const_76_promoted_to_fp16)[name = string("lower_cast_fp16")]; + bool top_interleave_0 = const()[name = string("top_interleave_0"), val = bool(false)]; + tensor top_cast_fp16 = concat(axis = var_3995, interleave = top_interleave_0, values = (var_4306_cast_fp16, var_378_to_fp16))[name = string("top_cast_fp16")]; + bool bottom_interleave_0 = const()[name = string("bottom_interleave_0"), val = bool(false)]; + tensor bottom_cast_fp16 = concat(axis = var_3995, interleave = bottom_interleave_0, values = (lower_cast_fp16, var_4309_cast_fp16))[name = string("bottom_cast_fp16")]; + bool inv_93_interleave_0 = const()[name = string("inv_93_interleave_0"), val = bool(false)]; + tensor inv_93_cast_fp16 = concat(axis = var_3974, interleave = inv_93_interleave_0, values = (top_cast_fp16, bottom_cast_fp16))[name = string("inv_93_cast_fp16")]; + tensor var_4322 = const()[name = string("op_4322"), val = tensor([32, 16, 64, 64])]; + tensor inv_cast_fp16 = reshape(shape = var_4322, x = inv_93_cast_fp16)[name = string("inv_cast_fp16")]; + bool new_v_transpose_x_0 = const()[name = string("new_v_transpose_x_0"), val = bool(false)]; + bool new_v_transpose_y_0 = const()[name = string("new_v_transpose_y_0"), val = bool(false)]; + tensor new_v_cast_fp16 = matmul(transpose_x = new_v_transpose_x_0, transpose_y = new_v_transpose_y_0, x = inv_cast_fp16, y = v_beta_cast_fp16)[name = string("new_v_cast_fp16")]; + tensor var_4325_cast_fp16 = exp(x = cum_cast_fp16)[name = string("op_4325_cast_fp16")]; + tensor var_4326_axes_0 = const()[name = string("op_4326_axes_0"), val = tensor([3])]; + tensor var_4326_cast_fp16 = expand_dims(axes = var_4326_axes_0, x = var_4325_cast_fp16)[name = string("op_4326_cast_fp16")]; + tensor var_4327_cast_fp16 = mul(x = k_beta_cast_fp16, y = var_4326_cast_fp16)[name = string("op_4327_cast_fp16")]; + bool k_cumdecay_transpose_x_0 = const()[name = string("k_cumdecay_transpose_x_0"), val = bool(false)]; + bool k_cumdecay_transpose_y_0 = const()[name = string("k_cumdecay_transpose_y_0"), val = bool(false)]; + tensor k_cumdecay_cast_fp16 = matmul(transpose_x = k_cumdecay_transpose_x_0, transpose_y = k_cumdecay_transpose_y_0, x = inv_cast_fp16, y = var_4327_cast_fp16)[name = string("k_cumdecay_cast_fp16")]; + tensor q_dec_cast_fp16 = mul(x = q_63_cast_fp16, y = var_4326_cast_fp16)[name = string("q_dec_cast_fp16")]; + tensor var_4332_cast_fp16 = exp(x = to_end_cast_fp16)[name = string("op_4332_cast_fp16")]; + tensor var_4333_axes_0 = const()[name = string("op_4333_axes_0"), val = tensor([3])]; + tensor var_4333_cast_fp16 = expand_dims(axes = var_4333_axes_0, x = var_4332_cast_fp16)[name = string("op_4333_cast_fp16")]; + tensor k_dec_cast_fp16 = mul(x = k_53_cast_fp16, y = var_4333_cast_fp16)[name = string("k_dec_cast_fp16")]; + tensor var_4335_begin_0 = const()[name = string("op_4335_begin_0"), val = tensor([0, 0, -1])]; + tensor var_4335_end_0 = const()[name = string("op_4335_end_0"), val = tensor([32, 16, 64])]; + tensor var_4335_end_mask_0 = const()[name = string("op_4335_end_mask_0"), val = tensor([true, true, true])]; + tensor var_4335_cast_fp16 = slice_by_index(begin = var_4335_begin_0, end = var_4335_end_0, end_mask = var_4335_end_mask_0, x = cum_cast_fp16)[name = string("op_4335_cast_fp16")]; + tensor var_4336_cast_fp16 = exp(x = var_4335_cast_fp16)[name = string("op_4336_cast_fp16")]; + tensor chunk_decay_axes_0 = const()[name = string("chunk_decay_axes_0"), val = tensor([3])]; + tensor chunk_decay_cast_fp16 = expand_dims(axes = chunk_decay_axes_0, x = var_4336_cast_fp16)[name = string("chunk_decay_cast_fp16")]; + tensor var_4341_begin_0 = const()[name = string("op_4341_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4341_end_0 = const()[name = string("op_4341_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_4341_end_mask_0 = const()[name = string("op_4341_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4341_squeeze_mask_0 = const()[name = string("op_4341_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4341_cast_fp16 = slice_by_index(begin = var_4341_begin_0, end = var_4341_end_0, end_mask = var_4341_end_mask_0, squeeze_mask = var_4341_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4341_cast_fp16")]; + tensor var_4343_begin_0 = const()[name = string("op_4343_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4343_end_0 = const()[name = string("op_4343_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_4343_end_mask_0 = const()[name = string("op_4343_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4343_squeeze_mask_0 = const()[name = string("op_4343_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4343_cast_fp16 = slice_by_index(begin = var_4343_begin_0, end = var_4343_end_0, end_mask = var_4343_end_mask_0, squeeze_mask = var_4343_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4343_cast_fp16")]; + bool var_4344_transpose_x_0 = const()[name = string("op_4344_transpose_x_0"), val = bool(false)]; + bool var_4344_transpose_y_0 = const()[name = string("op_4344_transpose_y_0"), val = bool(false)]; + tensor var_4344_cast_fp16 = matmul(transpose_x = var_4344_transpose_x_0, transpose_y = var_4344_transpose_y_0, x = var_4343_cast_fp16, y = state_1_to_fp16)[name = string("op_4344_cast_fp16")]; + tensor v_new_161_cast_fp16 = sub(x = var_4341_cast_fp16, y = var_4344_cast_fp16)[name = string("v_new_161_cast_fp16")]; + tensor var_4347_begin_0 = const()[name = string("op_4347_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4347_end_0 = const()[name = string("op_4347_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_4347_end_mask_0 = const()[name = string("op_4347_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4347_squeeze_mask_0 = const()[name = string("op_4347_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4347_cast_fp16 = slice_by_index(begin = var_4347_begin_0, end = var_4347_end_0, end_mask = var_4347_end_mask_0, squeeze_mask = var_4347_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4347_cast_fp16")]; + bool var_4348_transpose_x_0 = const()[name = string("op_4348_transpose_x_0"), val = bool(false)]; + bool var_4348_transpose_y_0 = const()[name = string("op_4348_transpose_y_0"), val = bool(false)]; + tensor var_4348_cast_fp16 = matmul(transpose_x = var_4348_transpose_x_0, transpose_y = var_4348_transpose_y_0, x = var_4347_cast_fp16, y = state_1_to_fp16)[name = string("op_4348_cast_fp16")]; + tensor var_4350_begin_0 = const()[name = string("op_4350_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4350_end_0 = const()[name = string("op_4350_end_0"), val = tensor([32, 1, 64, 64])]; + tensor var_4350_end_mask_0 = const()[name = string("op_4350_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4350_squeeze_mask_0 = const()[name = string("op_4350_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4350_cast_fp16 = slice_by_index(begin = var_4350_begin_0, end = var_4350_end_0, end_mask = var_4350_end_mask_0, squeeze_mask = var_4350_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4350_cast_fp16")]; + bool var_4351_transpose_x_0 = const()[name = string("op_4351_transpose_x_0"), val = bool(false)]; + bool var_4351_transpose_y_0 = const()[name = string("op_4351_transpose_y_0"), val = bool(false)]; + tensor var_4351_cast_fp16 = matmul(transpose_x = var_4351_transpose_x_0, transpose_y = var_4351_transpose_y_0, x = var_4350_cast_fp16, y = v_new_161_cast_fp16)[name = string("op_4351_cast_fp16")]; + tensor var_4352_cast_fp16 = add(x = var_4348_cast_fp16, y = var_4351_cast_fp16)[name = string("op_4352_cast_fp16")]; + tensor var_4354_begin_0 = const()[name = string("op_4354_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4354_end_0 = const()[name = string("op_4354_end_0"), val = tensor([32, 1, 1, 1])]; + tensor var_4354_end_mask_0 = const()[name = string("op_4354_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4354_squeeze_mask_0 = const()[name = string("op_4354_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4354_cast_fp16 = slice_by_index(begin = var_4354_begin_0, end = var_4354_end_0, end_mask = var_4354_end_mask_0, squeeze_mask = var_4354_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4354_cast_fp16")]; + tensor var_4355_cast_fp16 = mul(x = state_1_to_fp16, y = var_4354_cast_fp16)[name = string("op_4355_cast_fp16")]; + tensor var_4357_begin_0 = const()[name = string("op_4357_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4357_end_0 = const()[name = string("op_4357_end_0"), val = tensor([32, 1, 64, 128])]; + tensor var_4357_end_mask_0 = const()[name = string("op_4357_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4357_squeeze_mask_0 = const()[name = string("op_4357_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4357_cast_fp16 = slice_by_index(begin = var_4357_begin_0, end = var_4357_end_0, end_mask = var_4357_end_mask_0, squeeze_mask = var_4357_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4357_cast_fp16")]; + bool var_4359_transpose_x_1 = const()[name = string("op_4359_transpose_x_1"), val = bool(true)]; + bool var_4359_transpose_y_1 = const()[name = string("op_4359_transpose_y_1"), val = bool(false)]; + tensor var_4359_cast_fp16 = matmul(transpose_x = var_4359_transpose_x_1, transpose_y = var_4359_transpose_y_1, x = var_4357_cast_fp16, y = v_new_161_cast_fp16)[name = string("op_4359_cast_fp16")]; + tensor state_163_cast_fp16 = add(x = var_4355_cast_fp16, y = var_4359_cast_fp16)[name = string("state_163_cast_fp16")]; + tensor var_4362_begin_0 = const()[name = string("op_4362_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4362_end_0 = const()[name = string("op_4362_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_4362_end_mask_0 = const()[name = string("op_4362_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4362_squeeze_mask_0 = const()[name = string("op_4362_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4362_cast_fp16 = slice_by_index(begin = var_4362_begin_0, end = var_4362_end_0, end_mask = var_4362_end_mask_0, squeeze_mask = var_4362_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4362_cast_fp16")]; + tensor var_4364_begin_0 = const()[name = string("op_4364_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4364_end_0 = const()[name = string("op_4364_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_4364_end_mask_0 = const()[name = string("op_4364_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4364_squeeze_mask_0 = const()[name = string("op_4364_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4364_cast_fp16 = slice_by_index(begin = var_4364_begin_0, end = var_4364_end_0, end_mask = var_4364_end_mask_0, squeeze_mask = var_4364_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4364_cast_fp16")]; + bool var_4365_transpose_x_0 = const()[name = string("op_4365_transpose_x_0"), val = bool(false)]; + bool var_4365_transpose_y_0 = const()[name = string("op_4365_transpose_y_0"), val = bool(false)]; + tensor var_4365_cast_fp16 = matmul(transpose_x = var_4365_transpose_x_0, transpose_y = var_4365_transpose_y_0, x = var_4364_cast_fp16, y = state_163_cast_fp16)[name = string("op_4365_cast_fp16")]; + tensor v_new_163_cast_fp16 = sub(x = var_4362_cast_fp16, y = var_4365_cast_fp16)[name = string("v_new_163_cast_fp16")]; + tensor var_4368_begin_0 = const()[name = string("op_4368_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4368_end_0 = const()[name = string("op_4368_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_4368_end_mask_0 = const()[name = string("op_4368_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4368_squeeze_mask_0 = const()[name = string("op_4368_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4368_cast_fp16 = slice_by_index(begin = var_4368_begin_0, end = var_4368_end_0, end_mask = var_4368_end_mask_0, squeeze_mask = var_4368_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4368_cast_fp16")]; + bool var_4369_transpose_x_0 = const()[name = string("op_4369_transpose_x_0"), val = bool(false)]; + bool var_4369_transpose_y_0 = const()[name = string("op_4369_transpose_y_0"), val = bool(false)]; + tensor var_4369_cast_fp16 = matmul(transpose_x = var_4369_transpose_x_0, transpose_y = var_4369_transpose_y_0, x = var_4368_cast_fp16, y = state_163_cast_fp16)[name = string("op_4369_cast_fp16")]; + tensor var_4371_begin_0 = const()[name = string("op_4371_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4371_end_0 = const()[name = string("op_4371_end_0"), val = tensor([32, 2, 64, 64])]; + tensor var_4371_end_mask_0 = const()[name = string("op_4371_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4371_squeeze_mask_0 = const()[name = string("op_4371_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4371_cast_fp16 = slice_by_index(begin = var_4371_begin_0, end = var_4371_end_0, end_mask = var_4371_end_mask_0, squeeze_mask = var_4371_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4371_cast_fp16")]; + bool var_4372_transpose_x_0 = const()[name = string("op_4372_transpose_x_0"), val = bool(false)]; + bool var_4372_transpose_y_0 = const()[name = string("op_4372_transpose_y_0"), val = bool(false)]; + tensor var_4372_cast_fp16 = matmul(transpose_x = var_4372_transpose_x_0, transpose_y = var_4372_transpose_y_0, x = var_4371_cast_fp16, y = v_new_163_cast_fp16)[name = string("op_4372_cast_fp16")]; + tensor var_4373_cast_fp16 = add(x = var_4369_cast_fp16, y = var_4372_cast_fp16)[name = string("op_4373_cast_fp16")]; + tensor var_4375_begin_0 = const()[name = string("op_4375_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4375_end_0 = const()[name = string("op_4375_end_0"), val = tensor([32, 2, 1, 1])]; + tensor var_4375_end_mask_0 = const()[name = string("op_4375_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4375_squeeze_mask_0 = const()[name = string("op_4375_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4375_cast_fp16 = slice_by_index(begin = var_4375_begin_0, end = var_4375_end_0, end_mask = var_4375_end_mask_0, squeeze_mask = var_4375_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4375_cast_fp16")]; + tensor var_4376_cast_fp16 = mul(x = state_163_cast_fp16, y = var_4375_cast_fp16)[name = string("op_4376_cast_fp16")]; + tensor var_4378_begin_0 = const()[name = string("op_4378_begin_0"), val = tensor([0, 1, 0, 0])]; + tensor var_4378_end_0 = const()[name = string("op_4378_end_0"), val = tensor([32, 2, 64, 128])]; + tensor var_4378_end_mask_0 = const()[name = string("op_4378_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4378_squeeze_mask_0 = const()[name = string("op_4378_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4378_cast_fp16 = slice_by_index(begin = var_4378_begin_0, end = var_4378_end_0, end_mask = var_4378_end_mask_0, squeeze_mask = var_4378_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4378_cast_fp16")]; + bool var_4380_transpose_x_1 = const()[name = string("op_4380_transpose_x_1"), val = bool(true)]; + bool var_4380_transpose_y_1 = const()[name = string("op_4380_transpose_y_1"), val = bool(false)]; + tensor var_4380_cast_fp16 = matmul(transpose_x = var_4380_transpose_x_1, transpose_y = var_4380_transpose_y_1, x = var_4378_cast_fp16, y = v_new_163_cast_fp16)[name = string("op_4380_cast_fp16")]; + tensor state_165_cast_fp16 = add(x = var_4376_cast_fp16, y = var_4380_cast_fp16)[name = string("state_165_cast_fp16")]; + tensor var_4383_begin_0 = const()[name = string("op_4383_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4383_end_0 = const()[name = string("op_4383_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_4383_end_mask_0 = const()[name = string("op_4383_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4383_squeeze_mask_0 = const()[name = string("op_4383_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4383_cast_fp16 = slice_by_index(begin = var_4383_begin_0, end = var_4383_end_0, end_mask = var_4383_end_mask_0, squeeze_mask = var_4383_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4383_cast_fp16")]; + tensor var_4385_begin_0 = const()[name = string("op_4385_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4385_end_0 = const()[name = string("op_4385_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_4385_end_mask_0 = const()[name = string("op_4385_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4385_squeeze_mask_0 = const()[name = string("op_4385_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4385_cast_fp16 = slice_by_index(begin = var_4385_begin_0, end = var_4385_end_0, end_mask = var_4385_end_mask_0, squeeze_mask = var_4385_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4385_cast_fp16")]; + bool var_4386_transpose_x_0 = const()[name = string("op_4386_transpose_x_0"), val = bool(false)]; + bool var_4386_transpose_y_0 = const()[name = string("op_4386_transpose_y_0"), val = bool(false)]; + tensor var_4386_cast_fp16 = matmul(transpose_x = var_4386_transpose_x_0, transpose_y = var_4386_transpose_y_0, x = var_4385_cast_fp16, y = state_165_cast_fp16)[name = string("op_4386_cast_fp16")]; + tensor v_new_165_cast_fp16 = sub(x = var_4383_cast_fp16, y = var_4386_cast_fp16)[name = string("v_new_165_cast_fp16")]; + tensor var_4389_begin_0 = const()[name = string("op_4389_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4389_end_0 = const()[name = string("op_4389_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_4389_end_mask_0 = const()[name = string("op_4389_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4389_squeeze_mask_0 = const()[name = string("op_4389_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4389_cast_fp16 = slice_by_index(begin = var_4389_begin_0, end = var_4389_end_0, end_mask = var_4389_end_mask_0, squeeze_mask = var_4389_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4389_cast_fp16")]; + bool var_4390_transpose_x_0 = const()[name = string("op_4390_transpose_x_0"), val = bool(false)]; + bool var_4390_transpose_y_0 = const()[name = string("op_4390_transpose_y_0"), val = bool(false)]; + tensor var_4390_cast_fp16 = matmul(transpose_x = var_4390_transpose_x_0, transpose_y = var_4390_transpose_y_0, x = var_4389_cast_fp16, y = state_165_cast_fp16)[name = string("op_4390_cast_fp16")]; + tensor var_4392_begin_0 = const()[name = string("op_4392_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4392_end_0 = const()[name = string("op_4392_end_0"), val = tensor([32, 3, 64, 64])]; + tensor var_4392_end_mask_0 = const()[name = string("op_4392_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4392_squeeze_mask_0 = const()[name = string("op_4392_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4392_cast_fp16 = slice_by_index(begin = var_4392_begin_0, end = var_4392_end_0, end_mask = var_4392_end_mask_0, squeeze_mask = var_4392_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4392_cast_fp16")]; + bool var_4393_transpose_x_0 = const()[name = string("op_4393_transpose_x_0"), val = bool(false)]; + bool var_4393_transpose_y_0 = const()[name = string("op_4393_transpose_y_0"), val = bool(false)]; + tensor var_4393_cast_fp16 = matmul(transpose_x = var_4393_transpose_x_0, transpose_y = var_4393_transpose_y_0, x = var_4392_cast_fp16, y = v_new_165_cast_fp16)[name = string("op_4393_cast_fp16")]; + tensor var_4394_cast_fp16 = add(x = var_4390_cast_fp16, y = var_4393_cast_fp16)[name = string("op_4394_cast_fp16")]; + tensor var_4396_begin_0 = const()[name = string("op_4396_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4396_end_0 = const()[name = string("op_4396_end_0"), val = tensor([32, 3, 1, 1])]; + tensor var_4396_end_mask_0 = const()[name = string("op_4396_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4396_squeeze_mask_0 = const()[name = string("op_4396_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4396_cast_fp16 = slice_by_index(begin = var_4396_begin_0, end = var_4396_end_0, end_mask = var_4396_end_mask_0, squeeze_mask = var_4396_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4396_cast_fp16")]; + tensor var_4397_cast_fp16 = mul(x = state_165_cast_fp16, y = var_4396_cast_fp16)[name = string("op_4397_cast_fp16")]; + tensor var_4399_begin_0 = const()[name = string("op_4399_begin_0"), val = tensor([0, 2, 0, 0])]; + tensor var_4399_end_0 = const()[name = string("op_4399_end_0"), val = tensor([32, 3, 64, 128])]; + tensor var_4399_end_mask_0 = const()[name = string("op_4399_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4399_squeeze_mask_0 = const()[name = string("op_4399_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4399_cast_fp16 = slice_by_index(begin = var_4399_begin_0, end = var_4399_end_0, end_mask = var_4399_end_mask_0, squeeze_mask = var_4399_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4399_cast_fp16")]; + bool var_4401_transpose_x_1 = const()[name = string("op_4401_transpose_x_1"), val = bool(true)]; + bool var_4401_transpose_y_1 = const()[name = string("op_4401_transpose_y_1"), val = bool(false)]; + tensor var_4401_cast_fp16 = matmul(transpose_x = var_4401_transpose_x_1, transpose_y = var_4401_transpose_y_1, x = var_4399_cast_fp16, y = v_new_165_cast_fp16)[name = string("op_4401_cast_fp16")]; + tensor state_167_cast_fp16 = add(x = var_4397_cast_fp16, y = var_4401_cast_fp16)[name = string("state_167_cast_fp16")]; + tensor var_4404_begin_0 = const()[name = string("op_4404_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4404_end_0 = const()[name = string("op_4404_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_4404_end_mask_0 = const()[name = string("op_4404_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4404_squeeze_mask_0 = const()[name = string("op_4404_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4404_cast_fp16 = slice_by_index(begin = var_4404_begin_0, end = var_4404_end_0, end_mask = var_4404_end_mask_0, squeeze_mask = var_4404_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4404_cast_fp16")]; + tensor var_4406_begin_0 = const()[name = string("op_4406_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4406_end_0 = const()[name = string("op_4406_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_4406_end_mask_0 = const()[name = string("op_4406_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4406_squeeze_mask_0 = const()[name = string("op_4406_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4406_cast_fp16 = slice_by_index(begin = var_4406_begin_0, end = var_4406_end_0, end_mask = var_4406_end_mask_0, squeeze_mask = var_4406_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4406_cast_fp16")]; + bool var_4407_transpose_x_0 = const()[name = string("op_4407_transpose_x_0"), val = bool(false)]; + bool var_4407_transpose_y_0 = const()[name = string("op_4407_transpose_y_0"), val = bool(false)]; + tensor var_4407_cast_fp16 = matmul(transpose_x = var_4407_transpose_x_0, transpose_y = var_4407_transpose_y_0, x = var_4406_cast_fp16, y = state_167_cast_fp16)[name = string("op_4407_cast_fp16")]; + tensor v_new_167_cast_fp16 = sub(x = var_4404_cast_fp16, y = var_4407_cast_fp16)[name = string("v_new_167_cast_fp16")]; + tensor var_4410_begin_0 = const()[name = string("op_4410_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4410_end_0 = const()[name = string("op_4410_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_4410_end_mask_0 = const()[name = string("op_4410_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4410_squeeze_mask_0 = const()[name = string("op_4410_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4410_cast_fp16 = slice_by_index(begin = var_4410_begin_0, end = var_4410_end_0, end_mask = var_4410_end_mask_0, squeeze_mask = var_4410_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4410_cast_fp16")]; + bool var_4411_transpose_x_0 = const()[name = string("op_4411_transpose_x_0"), val = bool(false)]; + bool var_4411_transpose_y_0 = const()[name = string("op_4411_transpose_y_0"), val = bool(false)]; + tensor var_4411_cast_fp16 = matmul(transpose_x = var_4411_transpose_x_0, transpose_y = var_4411_transpose_y_0, x = var_4410_cast_fp16, y = state_167_cast_fp16)[name = string("op_4411_cast_fp16")]; + tensor var_4413_begin_0 = const()[name = string("op_4413_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4413_end_0 = const()[name = string("op_4413_end_0"), val = tensor([32, 4, 64, 64])]; + tensor var_4413_end_mask_0 = const()[name = string("op_4413_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4413_squeeze_mask_0 = const()[name = string("op_4413_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4413_cast_fp16 = slice_by_index(begin = var_4413_begin_0, end = var_4413_end_0, end_mask = var_4413_end_mask_0, squeeze_mask = var_4413_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4413_cast_fp16")]; + bool var_4414_transpose_x_0 = const()[name = string("op_4414_transpose_x_0"), val = bool(false)]; + bool var_4414_transpose_y_0 = const()[name = string("op_4414_transpose_y_0"), val = bool(false)]; + tensor var_4414_cast_fp16 = matmul(transpose_x = var_4414_transpose_x_0, transpose_y = var_4414_transpose_y_0, x = var_4413_cast_fp16, y = v_new_167_cast_fp16)[name = string("op_4414_cast_fp16")]; + tensor var_4415_cast_fp16 = add(x = var_4411_cast_fp16, y = var_4414_cast_fp16)[name = string("op_4415_cast_fp16")]; + tensor var_4417_begin_0 = const()[name = string("op_4417_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4417_end_0 = const()[name = string("op_4417_end_0"), val = tensor([32, 4, 1, 1])]; + tensor var_4417_end_mask_0 = const()[name = string("op_4417_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4417_squeeze_mask_0 = const()[name = string("op_4417_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4417_cast_fp16 = slice_by_index(begin = var_4417_begin_0, end = var_4417_end_0, end_mask = var_4417_end_mask_0, squeeze_mask = var_4417_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4417_cast_fp16")]; + tensor var_4418_cast_fp16 = mul(x = state_167_cast_fp16, y = var_4417_cast_fp16)[name = string("op_4418_cast_fp16")]; + tensor var_4420_begin_0 = const()[name = string("op_4420_begin_0"), val = tensor([0, 3, 0, 0])]; + tensor var_4420_end_0 = const()[name = string("op_4420_end_0"), val = tensor([32, 4, 64, 128])]; + tensor var_4420_end_mask_0 = const()[name = string("op_4420_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4420_squeeze_mask_0 = const()[name = string("op_4420_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4420_cast_fp16 = slice_by_index(begin = var_4420_begin_0, end = var_4420_end_0, end_mask = var_4420_end_mask_0, squeeze_mask = var_4420_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4420_cast_fp16")]; + bool var_4422_transpose_x_1 = const()[name = string("op_4422_transpose_x_1"), val = bool(true)]; + bool var_4422_transpose_y_1 = const()[name = string("op_4422_transpose_y_1"), val = bool(false)]; + tensor var_4422_cast_fp16 = matmul(transpose_x = var_4422_transpose_x_1, transpose_y = var_4422_transpose_y_1, x = var_4420_cast_fp16, y = v_new_167_cast_fp16)[name = string("op_4422_cast_fp16")]; + tensor state_169_cast_fp16 = add(x = var_4418_cast_fp16, y = var_4422_cast_fp16)[name = string("state_169_cast_fp16")]; + tensor var_4425_begin_0 = const()[name = string("op_4425_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4425_end_0 = const()[name = string("op_4425_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_4425_end_mask_0 = const()[name = string("op_4425_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4425_squeeze_mask_0 = const()[name = string("op_4425_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4425_cast_fp16 = slice_by_index(begin = var_4425_begin_0, end = var_4425_end_0, end_mask = var_4425_end_mask_0, squeeze_mask = var_4425_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4425_cast_fp16")]; + tensor var_4427_begin_0 = const()[name = string("op_4427_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4427_end_0 = const()[name = string("op_4427_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_4427_end_mask_0 = const()[name = string("op_4427_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4427_squeeze_mask_0 = const()[name = string("op_4427_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4427_cast_fp16 = slice_by_index(begin = var_4427_begin_0, end = var_4427_end_0, end_mask = var_4427_end_mask_0, squeeze_mask = var_4427_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4427_cast_fp16")]; + bool var_4428_transpose_x_0 = const()[name = string("op_4428_transpose_x_0"), val = bool(false)]; + bool var_4428_transpose_y_0 = const()[name = string("op_4428_transpose_y_0"), val = bool(false)]; + tensor var_4428_cast_fp16 = matmul(transpose_x = var_4428_transpose_x_0, transpose_y = var_4428_transpose_y_0, x = var_4427_cast_fp16, y = state_169_cast_fp16)[name = string("op_4428_cast_fp16")]; + tensor v_new_169_cast_fp16 = sub(x = var_4425_cast_fp16, y = var_4428_cast_fp16)[name = string("v_new_169_cast_fp16")]; + tensor var_4431_begin_0 = const()[name = string("op_4431_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4431_end_0 = const()[name = string("op_4431_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_4431_end_mask_0 = const()[name = string("op_4431_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4431_squeeze_mask_0 = const()[name = string("op_4431_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4431_cast_fp16 = slice_by_index(begin = var_4431_begin_0, end = var_4431_end_0, end_mask = var_4431_end_mask_0, squeeze_mask = var_4431_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4431_cast_fp16")]; + bool var_4432_transpose_x_0 = const()[name = string("op_4432_transpose_x_0"), val = bool(false)]; + bool var_4432_transpose_y_0 = const()[name = string("op_4432_transpose_y_0"), val = bool(false)]; + tensor var_4432_cast_fp16 = matmul(transpose_x = var_4432_transpose_x_0, transpose_y = var_4432_transpose_y_0, x = var_4431_cast_fp16, y = state_169_cast_fp16)[name = string("op_4432_cast_fp16")]; + tensor var_4434_begin_0 = const()[name = string("op_4434_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4434_end_0 = const()[name = string("op_4434_end_0"), val = tensor([32, 5, 64, 64])]; + tensor var_4434_end_mask_0 = const()[name = string("op_4434_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4434_squeeze_mask_0 = const()[name = string("op_4434_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4434_cast_fp16 = slice_by_index(begin = var_4434_begin_0, end = var_4434_end_0, end_mask = var_4434_end_mask_0, squeeze_mask = var_4434_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4434_cast_fp16")]; + bool var_4435_transpose_x_0 = const()[name = string("op_4435_transpose_x_0"), val = bool(false)]; + bool var_4435_transpose_y_0 = const()[name = string("op_4435_transpose_y_0"), val = bool(false)]; + tensor var_4435_cast_fp16 = matmul(transpose_x = var_4435_transpose_x_0, transpose_y = var_4435_transpose_y_0, x = var_4434_cast_fp16, y = v_new_169_cast_fp16)[name = string("op_4435_cast_fp16")]; + tensor var_4436_cast_fp16 = add(x = var_4432_cast_fp16, y = var_4435_cast_fp16)[name = string("op_4436_cast_fp16")]; + tensor var_4438_begin_0 = const()[name = string("op_4438_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4438_end_0 = const()[name = string("op_4438_end_0"), val = tensor([32, 5, 1, 1])]; + tensor var_4438_end_mask_0 = const()[name = string("op_4438_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4438_squeeze_mask_0 = const()[name = string("op_4438_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4438_cast_fp16 = slice_by_index(begin = var_4438_begin_0, end = var_4438_end_0, end_mask = var_4438_end_mask_0, squeeze_mask = var_4438_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4438_cast_fp16")]; + tensor var_4439_cast_fp16 = mul(x = state_169_cast_fp16, y = var_4438_cast_fp16)[name = string("op_4439_cast_fp16")]; + tensor var_4441_begin_0 = const()[name = string("op_4441_begin_0"), val = tensor([0, 4, 0, 0])]; + tensor var_4441_end_0 = const()[name = string("op_4441_end_0"), val = tensor([32, 5, 64, 128])]; + tensor var_4441_end_mask_0 = const()[name = string("op_4441_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4441_squeeze_mask_0 = const()[name = string("op_4441_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4441_cast_fp16 = slice_by_index(begin = var_4441_begin_0, end = var_4441_end_0, end_mask = var_4441_end_mask_0, squeeze_mask = var_4441_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4441_cast_fp16")]; + bool var_4443_transpose_x_1 = const()[name = string("op_4443_transpose_x_1"), val = bool(true)]; + bool var_4443_transpose_y_1 = const()[name = string("op_4443_transpose_y_1"), val = bool(false)]; + tensor var_4443_cast_fp16 = matmul(transpose_x = var_4443_transpose_x_1, transpose_y = var_4443_transpose_y_1, x = var_4441_cast_fp16, y = v_new_169_cast_fp16)[name = string("op_4443_cast_fp16")]; + tensor state_171_cast_fp16 = add(x = var_4439_cast_fp16, y = var_4443_cast_fp16)[name = string("state_171_cast_fp16")]; + tensor var_4446_begin_0 = const()[name = string("op_4446_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4446_end_0 = const()[name = string("op_4446_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_4446_end_mask_0 = const()[name = string("op_4446_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4446_squeeze_mask_0 = const()[name = string("op_4446_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4446_cast_fp16 = slice_by_index(begin = var_4446_begin_0, end = var_4446_end_0, end_mask = var_4446_end_mask_0, squeeze_mask = var_4446_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4446_cast_fp16")]; + tensor var_4448_begin_0 = const()[name = string("op_4448_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4448_end_0 = const()[name = string("op_4448_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_4448_end_mask_0 = const()[name = string("op_4448_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4448_squeeze_mask_0 = const()[name = string("op_4448_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4448_cast_fp16 = slice_by_index(begin = var_4448_begin_0, end = var_4448_end_0, end_mask = var_4448_end_mask_0, squeeze_mask = var_4448_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4448_cast_fp16")]; + bool var_4449_transpose_x_0 = const()[name = string("op_4449_transpose_x_0"), val = bool(false)]; + bool var_4449_transpose_y_0 = const()[name = string("op_4449_transpose_y_0"), val = bool(false)]; + tensor var_4449_cast_fp16 = matmul(transpose_x = var_4449_transpose_x_0, transpose_y = var_4449_transpose_y_0, x = var_4448_cast_fp16, y = state_171_cast_fp16)[name = string("op_4449_cast_fp16")]; + tensor v_new_171_cast_fp16 = sub(x = var_4446_cast_fp16, y = var_4449_cast_fp16)[name = string("v_new_171_cast_fp16")]; + tensor var_4452_begin_0 = const()[name = string("op_4452_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4452_end_0 = const()[name = string("op_4452_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_4452_end_mask_0 = const()[name = string("op_4452_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4452_squeeze_mask_0 = const()[name = string("op_4452_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4452_cast_fp16 = slice_by_index(begin = var_4452_begin_0, end = var_4452_end_0, end_mask = var_4452_end_mask_0, squeeze_mask = var_4452_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4452_cast_fp16")]; + bool var_4453_transpose_x_0 = const()[name = string("op_4453_transpose_x_0"), val = bool(false)]; + bool var_4453_transpose_y_0 = const()[name = string("op_4453_transpose_y_0"), val = bool(false)]; + tensor var_4453_cast_fp16 = matmul(transpose_x = var_4453_transpose_x_0, transpose_y = var_4453_transpose_y_0, x = var_4452_cast_fp16, y = state_171_cast_fp16)[name = string("op_4453_cast_fp16")]; + tensor var_4455_begin_0 = const()[name = string("op_4455_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4455_end_0 = const()[name = string("op_4455_end_0"), val = tensor([32, 6, 64, 64])]; + tensor var_4455_end_mask_0 = const()[name = string("op_4455_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4455_squeeze_mask_0 = const()[name = string("op_4455_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4455_cast_fp16 = slice_by_index(begin = var_4455_begin_0, end = var_4455_end_0, end_mask = var_4455_end_mask_0, squeeze_mask = var_4455_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4455_cast_fp16")]; + bool var_4456_transpose_x_0 = const()[name = string("op_4456_transpose_x_0"), val = bool(false)]; + bool var_4456_transpose_y_0 = const()[name = string("op_4456_transpose_y_0"), val = bool(false)]; + tensor var_4456_cast_fp16 = matmul(transpose_x = var_4456_transpose_x_0, transpose_y = var_4456_transpose_y_0, x = var_4455_cast_fp16, y = v_new_171_cast_fp16)[name = string("op_4456_cast_fp16")]; + tensor var_4457_cast_fp16 = add(x = var_4453_cast_fp16, y = var_4456_cast_fp16)[name = string("op_4457_cast_fp16")]; + tensor var_4459_begin_0 = const()[name = string("op_4459_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4459_end_0 = const()[name = string("op_4459_end_0"), val = tensor([32, 6, 1, 1])]; + tensor var_4459_end_mask_0 = const()[name = string("op_4459_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4459_squeeze_mask_0 = const()[name = string("op_4459_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4459_cast_fp16 = slice_by_index(begin = var_4459_begin_0, end = var_4459_end_0, end_mask = var_4459_end_mask_0, squeeze_mask = var_4459_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4459_cast_fp16")]; + tensor var_4460_cast_fp16 = mul(x = state_171_cast_fp16, y = var_4459_cast_fp16)[name = string("op_4460_cast_fp16")]; + tensor var_4462_begin_0 = const()[name = string("op_4462_begin_0"), val = tensor([0, 5, 0, 0])]; + tensor var_4462_end_0 = const()[name = string("op_4462_end_0"), val = tensor([32, 6, 64, 128])]; + tensor var_4462_end_mask_0 = const()[name = string("op_4462_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4462_squeeze_mask_0 = const()[name = string("op_4462_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4462_cast_fp16 = slice_by_index(begin = var_4462_begin_0, end = var_4462_end_0, end_mask = var_4462_end_mask_0, squeeze_mask = var_4462_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4462_cast_fp16")]; + bool var_4464_transpose_x_1 = const()[name = string("op_4464_transpose_x_1"), val = bool(true)]; + bool var_4464_transpose_y_1 = const()[name = string("op_4464_transpose_y_1"), val = bool(false)]; + tensor var_4464_cast_fp16 = matmul(transpose_x = var_4464_transpose_x_1, transpose_y = var_4464_transpose_y_1, x = var_4462_cast_fp16, y = v_new_171_cast_fp16)[name = string("op_4464_cast_fp16")]; + tensor state_173_cast_fp16 = add(x = var_4460_cast_fp16, y = var_4464_cast_fp16)[name = string("state_173_cast_fp16")]; + tensor var_4467_begin_0 = const()[name = string("op_4467_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4467_end_0 = const()[name = string("op_4467_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_4467_end_mask_0 = const()[name = string("op_4467_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4467_squeeze_mask_0 = const()[name = string("op_4467_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4467_cast_fp16 = slice_by_index(begin = var_4467_begin_0, end = var_4467_end_0, end_mask = var_4467_end_mask_0, squeeze_mask = var_4467_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4467_cast_fp16")]; + tensor var_4469_begin_0 = const()[name = string("op_4469_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4469_end_0 = const()[name = string("op_4469_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_4469_end_mask_0 = const()[name = string("op_4469_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4469_squeeze_mask_0 = const()[name = string("op_4469_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4469_cast_fp16 = slice_by_index(begin = var_4469_begin_0, end = var_4469_end_0, end_mask = var_4469_end_mask_0, squeeze_mask = var_4469_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4469_cast_fp16")]; + bool var_4470_transpose_x_0 = const()[name = string("op_4470_transpose_x_0"), val = bool(false)]; + bool var_4470_transpose_y_0 = const()[name = string("op_4470_transpose_y_0"), val = bool(false)]; + tensor var_4470_cast_fp16 = matmul(transpose_x = var_4470_transpose_x_0, transpose_y = var_4470_transpose_y_0, x = var_4469_cast_fp16, y = state_173_cast_fp16)[name = string("op_4470_cast_fp16")]; + tensor v_new_173_cast_fp16 = sub(x = var_4467_cast_fp16, y = var_4470_cast_fp16)[name = string("v_new_173_cast_fp16")]; + tensor var_4473_begin_0 = const()[name = string("op_4473_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4473_end_0 = const()[name = string("op_4473_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_4473_end_mask_0 = const()[name = string("op_4473_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4473_squeeze_mask_0 = const()[name = string("op_4473_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4473_cast_fp16 = slice_by_index(begin = var_4473_begin_0, end = var_4473_end_0, end_mask = var_4473_end_mask_0, squeeze_mask = var_4473_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4473_cast_fp16")]; + bool var_4474_transpose_x_0 = const()[name = string("op_4474_transpose_x_0"), val = bool(false)]; + bool var_4474_transpose_y_0 = const()[name = string("op_4474_transpose_y_0"), val = bool(false)]; + tensor var_4474_cast_fp16 = matmul(transpose_x = var_4474_transpose_x_0, transpose_y = var_4474_transpose_y_0, x = var_4473_cast_fp16, y = state_173_cast_fp16)[name = string("op_4474_cast_fp16")]; + tensor var_4476_begin_0 = const()[name = string("op_4476_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4476_end_0 = const()[name = string("op_4476_end_0"), val = tensor([32, 7, 64, 64])]; + tensor var_4476_end_mask_0 = const()[name = string("op_4476_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4476_squeeze_mask_0 = const()[name = string("op_4476_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4476_cast_fp16 = slice_by_index(begin = var_4476_begin_0, end = var_4476_end_0, end_mask = var_4476_end_mask_0, squeeze_mask = var_4476_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4476_cast_fp16")]; + bool var_4477_transpose_x_0 = const()[name = string("op_4477_transpose_x_0"), val = bool(false)]; + bool var_4477_transpose_y_0 = const()[name = string("op_4477_transpose_y_0"), val = bool(false)]; + tensor var_4477_cast_fp16 = matmul(transpose_x = var_4477_transpose_x_0, transpose_y = var_4477_transpose_y_0, x = var_4476_cast_fp16, y = v_new_173_cast_fp16)[name = string("op_4477_cast_fp16")]; + tensor var_4478_cast_fp16 = add(x = var_4474_cast_fp16, y = var_4477_cast_fp16)[name = string("op_4478_cast_fp16")]; + tensor var_4480_begin_0 = const()[name = string("op_4480_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4480_end_0 = const()[name = string("op_4480_end_0"), val = tensor([32, 7, 1, 1])]; + tensor var_4480_end_mask_0 = const()[name = string("op_4480_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4480_squeeze_mask_0 = const()[name = string("op_4480_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4480_cast_fp16 = slice_by_index(begin = var_4480_begin_0, end = var_4480_end_0, end_mask = var_4480_end_mask_0, squeeze_mask = var_4480_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4480_cast_fp16")]; + tensor var_4481_cast_fp16 = mul(x = state_173_cast_fp16, y = var_4480_cast_fp16)[name = string("op_4481_cast_fp16")]; + tensor var_4483_begin_0 = const()[name = string("op_4483_begin_0"), val = tensor([0, 6, 0, 0])]; + tensor var_4483_end_0 = const()[name = string("op_4483_end_0"), val = tensor([32, 7, 64, 128])]; + tensor var_4483_end_mask_0 = const()[name = string("op_4483_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4483_squeeze_mask_0 = const()[name = string("op_4483_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4483_cast_fp16 = slice_by_index(begin = var_4483_begin_0, end = var_4483_end_0, end_mask = var_4483_end_mask_0, squeeze_mask = var_4483_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4483_cast_fp16")]; + bool var_4485_transpose_x_1 = const()[name = string("op_4485_transpose_x_1"), val = bool(true)]; + bool var_4485_transpose_y_1 = const()[name = string("op_4485_transpose_y_1"), val = bool(false)]; + tensor var_4485_cast_fp16 = matmul(transpose_x = var_4485_transpose_x_1, transpose_y = var_4485_transpose_y_1, x = var_4483_cast_fp16, y = v_new_173_cast_fp16)[name = string("op_4485_cast_fp16")]; + tensor state_175_cast_fp16 = add(x = var_4481_cast_fp16, y = var_4485_cast_fp16)[name = string("state_175_cast_fp16")]; + tensor var_4488_begin_0 = const()[name = string("op_4488_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4488_end_0 = const()[name = string("op_4488_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_4488_end_mask_0 = const()[name = string("op_4488_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4488_squeeze_mask_0 = const()[name = string("op_4488_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4488_cast_fp16 = slice_by_index(begin = var_4488_begin_0, end = var_4488_end_0, end_mask = var_4488_end_mask_0, squeeze_mask = var_4488_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4488_cast_fp16")]; + tensor var_4490_begin_0 = const()[name = string("op_4490_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4490_end_0 = const()[name = string("op_4490_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_4490_end_mask_0 = const()[name = string("op_4490_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4490_squeeze_mask_0 = const()[name = string("op_4490_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4490_cast_fp16 = slice_by_index(begin = var_4490_begin_0, end = var_4490_end_0, end_mask = var_4490_end_mask_0, squeeze_mask = var_4490_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4490_cast_fp16")]; + bool var_4491_transpose_x_0 = const()[name = string("op_4491_transpose_x_0"), val = bool(false)]; + bool var_4491_transpose_y_0 = const()[name = string("op_4491_transpose_y_0"), val = bool(false)]; + tensor var_4491_cast_fp16 = matmul(transpose_x = var_4491_transpose_x_0, transpose_y = var_4491_transpose_y_0, x = var_4490_cast_fp16, y = state_175_cast_fp16)[name = string("op_4491_cast_fp16")]; + tensor v_new_175_cast_fp16 = sub(x = var_4488_cast_fp16, y = var_4491_cast_fp16)[name = string("v_new_175_cast_fp16")]; + tensor var_4494_begin_0 = const()[name = string("op_4494_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4494_end_0 = const()[name = string("op_4494_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_4494_end_mask_0 = const()[name = string("op_4494_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4494_squeeze_mask_0 = const()[name = string("op_4494_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4494_cast_fp16 = slice_by_index(begin = var_4494_begin_0, end = var_4494_end_0, end_mask = var_4494_end_mask_0, squeeze_mask = var_4494_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4494_cast_fp16")]; + bool var_4495_transpose_x_0 = const()[name = string("op_4495_transpose_x_0"), val = bool(false)]; + bool var_4495_transpose_y_0 = const()[name = string("op_4495_transpose_y_0"), val = bool(false)]; + tensor var_4495_cast_fp16 = matmul(transpose_x = var_4495_transpose_x_0, transpose_y = var_4495_transpose_y_0, x = var_4494_cast_fp16, y = state_175_cast_fp16)[name = string("op_4495_cast_fp16")]; + tensor var_4497_begin_0 = const()[name = string("op_4497_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4497_end_0 = const()[name = string("op_4497_end_0"), val = tensor([32, 8, 64, 64])]; + tensor var_4497_end_mask_0 = const()[name = string("op_4497_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4497_squeeze_mask_0 = const()[name = string("op_4497_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4497_cast_fp16 = slice_by_index(begin = var_4497_begin_0, end = var_4497_end_0, end_mask = var_4497_end_mask_0, squeeze_mask = var_4497_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4497_cast_fp16")]; + bool var_4498_transpose_x_0 = const()[name = string("op_4498_transpose_x_0"), val = bool(false)]; + bool var_4498_transpose_y_0 = const()[name = string("op_4498_transpose_y_0"), val = bool(false)]; + tensor var_4498_cast_fp16 = matmul(transpose_x = var_4498_transpose_x_0, transpose_y = var_4498_transpose_y_0, x = var_4497_cast_fp16, y = v_new_175_cast_fp16)[name = string("op_4498_cast_fp16")]; + tensor var_4499_cast_fp16 = add(x = var_4495_cast_fp16, y = var_4498_cast_fp16)[name = string("op_4499_cast_fp16")]; + tensor var_4501_begin_0 = const()[name = string("op_4501_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4501_end_0 = const()[name = string("op_4501_end_0"), val = tensor([32, 8, 1, 1])]; + tensor var_4501_end_mask_0 = const()[name = string("op_4501_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4501_squeeze_mask_0 = const()[name = string("op_4501_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4501_cast_fp16 = slice_by_index(begin = var_4501_begin_0, end = var_4501_end_0, end_mask = var_4501_end_mask_0, squeeze_mask = var_4501_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4501_cast_fp16")]; + tensor var_4502_cast_fp16 = mul(x = state_175_cast_fp16, y = var_4501_cast_fp16)[name = string("op_4502_cast_fp16")]; + tensor var_4504_begin_0 = const()[name = string("op_4504_begin_0"), val = tensor([0, 7, 0, 0])]; + tensor var_4504_end_0 = const()[name = string("op_4504_end_0"), val = tensor([32, 8, 64, 128])]; + tensor var_4504_end_mask_0 = const()[name = string("op_4504_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4504_squeeze_mask_0 = const()[name = string("op_4504_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4504_cast_fp16 = slice_by_index(begin = var_4504_begin_0, end = var_4504_end_0, end_mask = var_4504_end_mask_0, squeeze_mask = var_4504_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4504_cast_fp16")]; + bool var_4506_transpose_x_1 = const()[name = string("op_4506_transpose_x_1"), val = bool(true)]; + bool var_4506_transpose_y_1 = const()[name = string("op_4506_transpose_y_1"), val = bool(false)]; + tensor var_4506_cast_fp16 = matmul(transpose_x = var_4506_transpose_x_1, transpose_y = var_4506_transpose_y_1, x = var_4504_cast_fp16, y = v_new_175_cast_fp16)[name = string("op_4506_cast_fp16")]; + tensor state_177_cast_fp16 = add(x = var_4502_cast_fp16, y = var_4506_cast_fp16)[name = string("state_177_cast_fp16")]; + tensor var_4509_begin_0 = const()[name = string("op_4509_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4509_end_0 = const()[name = string("op_4509_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_4509_end_mask_0 = const()[name = string("op_4509_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4509_squeeze_mask_0 = const()[name = string("op_4509_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4509_cast_fp16 = slice_by_index(begin = var_4509_begin_0, end = var_4509_end_0, end_mask = var_4509_end_mask_0, squeeze_mask = var_4509_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4509_cast_fp16")]; + tensor var_4511_begin_0 = const()[name = string("op_4511_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4511_end_0 = const()[name = string("op_4511_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_4511_end_mask_0 = const()[name = string("op_4511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4511_squeeze_mask_0 = const()[name = string("op_4511_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4511_cast_fp16 = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, squeeze_mask = var_4511_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4511_cast_fp16")]; + bool var_4512_transpose_x_0 = const()[name = string("op_4512_transpose_x_0"), val = bool(false)]; + bool var_4512_transpose_y_0 = const()[name = string("op_4512_transpose_y_0"), val = bool(false)]; + tensor var_4512_cast_fp16 = matmul(transpose_x = var_4512_transpose_x_0, transpose_y = var_4512_transpose_y_0, x = var_4511_cast_fp16, y = state_177_cast_fp16)[name = string("op_4512_cast_fp16")]; + tensor v_new_177_cast_fp16 = sub(x = var_4509_cast_fp16, y = var_4512_cast_fp16)[name = string("v_new_177_cast_fp16")]; + tensor var_4515_begin_0 = const()[name = string("op_4515_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4515_end_0 = const()[name = string("op_4515_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_4515_end_mask_0 = const()[name = string("op_4515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4515_squeeze_mask_0 = const()[name = string("op_4515_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4515_cast_fp16 = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, squeeze_mask = var_4515_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4515_cast_fp16")]; + bool var_4516_transpose_x_0 = const()[name = string("op_4516_transpose_x_0"), val = bool(false)]; + bool var_4516_transpose_y_0 = const()[name = string("op_4516_transpose_y_0"), val = bool(false)]; + tensor var_4516_cast_fp16 = matmul(transpose_x = var_4516_transpose_x_0, transpose_y = var_4516_transpose_y_0, x = var_4515_cast_fp16, y = state_177_cast_fp16)[name = string("op_4516_cast_fp16")]; + tensor var_4518_begin_0 = const()[name = string("op_4518_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4518_end_0 = const()[name = string("op_4518_end_0"), val = tensor([32, 9, 64, 64])]; + tensor var_4518_end_mask_0 = const()[name = string("op_4518_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4518_squeeze_mask_0 = const()[name = string("op_4518_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4518_cast_fp16 = slice_by_index(begin = var_4518_begin_0, end = var_4518_end_0, end_mask = var_4518_end_mask_0, squeeze_mask = var_4518_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4518_cast_fp16")]; + bool var_4519_transpose_x_0 = const()[name = string("op_4519_transpose_x_0"), val = bool(false)]; + bool var_4519_transpose_y_0 = const()[name = string("op_4519_transpose_y_0"), val = bool(false)]; + tensor var_4519_cast_fp16 = matmul(transpose_x = var_4519_transpose_x_0, transpose_y = var_4519_transpose_y_0, x = var_4518_cast_fp16, y = v_new_177_cast_fp16)[name = string("op_4519_cast_fp16")]; + tensor var_4520_cast_fp16 = add(x = var_4516_cast_fp16, y = var_4519_cast_fp16)[name = string("op_4520_cast_fp16")]; + tensor var_4522_begin_0 = const()[name = string("op_4522_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4522_end_0 = const()[name = string("op_4522_end_0"), val = tensor([32, 9, 1, 1])]; + tensor var_4522_end_mask_0 = const()[name = string("op_4522_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4522_squeeze_mask_0 = const()[name = string("op_4522_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4522_cast_fp16 = slice_by_index(begin = var_4522_begin_0, end = var_4522_end_0, end_mask = var_4522_end_mask_0, squeeze_mask = var_4522_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4522_cast_fp16")]; + tensor var_4523_cast_fp16 = mul(x = state_177_cast_fp16, y = var_4522_cast_fp16)[name = string("op_4523_cast_fp16")]; + tensor var_4525_begin_0 = const()[name = string("op_4525_begin_0"), val = tensor([0, 8, 0, 0])]; + tensor var_4525_end_0 = const()[name = string("op_4525_end_0"), val = tensor([32, 9, 64, 128])]; + tensor var_4525_end_mask_0 = const()[name = string("op_4525_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4525_squeeze_mask_0 = const()[name = string("op_4525_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4525_cast_fp16 = slice_by_index(begin = var_4525_begin_0, end = var_4525_end_0, end_mask = var_4525_end_mask_0, squeeze_mask = var_4525_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4525_cast_fp16")]; + bool var_4527_transpose_x_1 = const()[name = string("op_4527_transpose_x_1"), val = bool(true)]; + bool var_4527_transpose_y_1 = const()[name = string("op_4527_transpose_y_1"), val = bool(false)]; + tensor var_4527_cast_fp16 = matmul(transpose_x = var_4527_transpose_x_1, transpose_y = var_4527_transpose_y_1, x = var_4525_cast_fp16, y = v_new_177_cast_fp16)[name = string("op_4527_cast_fp16")]; + tensor state_179_cast_fp16 = add(x = var_4523_cast_fp16, y = var_4527_cast_fp16)[name = string("state_179_cast_fp16")]; + tensor var_4530_begin_0 = const()[name = string("op_4530_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4530_end_0 = const()[name = string("op_4530_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_4530_end_mask_0 = const()[name = string("op_4530_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4530_squeeze_mask_0 = const()[name = string("op_4530_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4530_cast_fp16 = slice_by_index(begin = var_4530_begin_0, end = var_4530_end_0, end_mask = var_4530_end_mask_0, squeeze_mask = var_4530_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4530_cast_fp16")]; + tensor var_4532_begin_0 = const()[name = string("op_4532_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4532_end_0 = const()[name = string("op_4532_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_4532_end_mask_0 = const()[name = string("op_4532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4532_squeeze_mask_0 = const()[name = string("op_4532_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4532_cast_fp16 = slice_by_index(begin = var_4532_begin_0, end = var_4532_end_0, end_mask = var_4532_end_mask_0, squeeze_mask = var_4532_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4532_cast_fp16")]; + bool var_4533_transpose_x_0 = const()[name = string("op_4533_transpose_x_0"), val = bool(false)]; + bool var_4533_transpose_y_0 = const()[name = string("op_4533_transpose_y_0"), val = bool(false)]; + tensor var_4533_cast_fp16 = matmul(transpose_x = var_4533_transpose_x_0, transpose_y = var_4533_transpose_y_0, x = var_4532_cast_fp16, y = state_179_cast_fp16)[name = string("op_4533_cast_fp16")]; + tensor v_new_179_cast_fp16 = sub(x = var_4530_cast_fp16, y = var_4533_cast_fp16)[name = string("v_new_179_cast_fp16")]; + tensor var_4536_begin_0 = const()[name = string("op_4536_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4536_end_0 = const()[name = string("op_4536_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_4536_end_mask_0 = const()[name = string("op_4536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4536_squeeze_mask_0 = const()[name = string("op_4536_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4536_cast_fp16 = slice_by_index(begin = var_4536_begin_0, end = var_4536_end_0, end_mask = var_4536_end_mask_0, squeeze_mask = var_4536_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4536_cast_fp16")]; + bool var_4537_transpose_x_0 = const()[name = string("op_4537_transpose_x_0"), val = bool(false)]; + bool var_4537_transpose_y_0 = const()[name = string("op_4537_transpose_y_0"), val = bool(false)]; + tensor var_4537_cast_fp16 = matmul(transpose_x = var_4537_transpose_x_0, transpose_y = var_4537_transpose_y_0, x = var_4536_cast_fp16, y = state_179_cast_fp16)[name = string("op_4537_cast_fp16")]; + tensor var_4539_begin_0 = const()[name = string("op_4539_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4539_end_0 = const()[name = string("op_4539_end_0"), val = tensor([32, 10, 64, 64])]; + tensor var_4539_end_mask_0 = const()[name = string("op_4539_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4539_squeeze_mask_0 = const()[name = string("op_4539_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4539_cast_fp16 = slice_by_index(begin = var_4539_begin_0, end = var_4539_end_0, end_mask = var_4539_end_mask_0, squeeze_mask = var_4539_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4539_cast_fp16")]; + bool var_4540_transpose_x_0 = const()[name = string("op_4540_transpose_x_0"), val = bool(false)]; + bool var_4540_transpose_y_0 = const()[name = string("op_4540_transpose_y_0"), val = bool(false)]; + tensor var_4540_cast_fp16 = matmul(transpose_x = var_4540_transpose_x_0, transpose_y = var_4540_transpose_y_0, x = var_4539_cast_fp16, y = v_new_179_cast_fp16)[name = string("op_4540_cast_fp16")]; + tensor var_4541_cast_fp16 = add(x = var_4537_cast_fp16, y = var_4540_cast_fp16)[name = string("op_4541_cast_fp16")]; + tensor var_4543_begin_0 = const()[name = string("op_4543_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4543_end_0 = const()[name = string("op_4543_end_0"), val = tensor([32, 10, 1, 1])]; + tensor var_4543_end_mask_0 = const()[name = string("op_4543_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4543_squeeze_mask_0 = const()[name = string("op_4543_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4543_cast_fp16 = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, squeeze_mask = var_4543_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4543_cast_fp16")]; + tensor var_4544_cast_fp16 = mul(x = state_179_cast_fp16, y = var_4543_cast_fp16)[name = string("op_4544_cast_fp16")]; + tensor var_4546_begin_0 = const()[name = string("op_4546_begin_0"), val = tensor([0, 9, 0, 0])]; + tensor var_4546_end_0 = const()[name = string("op_4546_end_0"), val = tensor([32, 10, 64, 128])]; + tensor var_4546_end_mask_0 = const()[name = string("op_4546_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4546_squeeze_mask_0 = const()[name = string("op_4546_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4546_cast_fp16 = slice_by_index(begin = var_4546_begin_0, end = var_4546_end_0, end_mask = var_4546_end_mask_0, squeeze_mask = var_4546_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4546_cast_fp16")]; + bool var_4548_transpose_x_1 = const()[name = string("op_4548_transpose_x_1"), val = bool(true)]; + bool var_4548_transpose_y_1 = const()[name = string("op_4548_transpose_y_1"), val = bool(false)]; + tensor var_4548_cast_fp16 = matmul(transpose_x = var_4548_transpose_x_1, transpose_y = var_4548_transpose_y_1, x = var_4546_cast_fp16, y = v_new_179_cast_fp16)[name = string("op_4548_cast_fp16")]; + tensor state_181_cast_fp16 = add(x = var_4544_cast_fp16, y = var_4548_cast_fp16)[name = string("state_181_cast_fp16")]; + tensor var_4551_begin_0 = const()[name = string("op_4551_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4551_end_0 = const()[name = string("op_4551_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_4551_end_mask_0 = const()[name = string("op_4551_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4551_squeeze_mask_0 = const()[name = string("op_4551_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4551_cast_fp16 = slice_by_index(begin = var_4551_begin_0, end = var_4551_end_0, end_mask = var_4551_end_mask_0, squeeze_mask = var_4551_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4551_cast_fp16")]; + tensor var_4553_begin_0 = const()[name = string("op_4553_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4553_end_0 = const()[name = string("op_4553_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_4553_end_mask_0 = const()[name = string("op_4553_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4553_squeeze_mask_0 = const()[name = string("op_4553_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4553_cast_fp16 = slice_by_index(begin = var_4553_begin_0, end = var_4553_end_0, end_mask = var_4553_end_mask_0, squeeze_mask = var_4553_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4553_cast_fp16")]; + bool var_4554_transpose_x_0 = const()[name = string("op_4554_transpose_x_0"), val = bool(false)]; + bool var_4554_transpose_y_0 = const()[name = string("op_4554_transpose_y_0"), val = bool(false)]; + tensor var_4554_cast_fp16 = matmul(transpose_x = var_4554_transpose_x_0, transpose_y = var_4554_transpose_y_0, x = var_4553_cast_fp16, y = state_181_cast_fp16)[name = string("op_4554_cast_fp16")]; + tensor v_new_181_cast_fp16 = sub(x = var_4551_cast_fp16, y = var_4554_cast_fp16)[name = string("v_new_181_cast_fp16")]; + tensor var_4557_begin_0 = const()[name = string("op_4557_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4557_end_0 = const()[name = string("op_4557_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_4557_end_mask_0 = const()[name = string("op_4557_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4557_squeeze_mask_0 = const()[name = string("op_4557_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4557_cast_fp16 = slice_by_index(begin = var_4557_begin_0, end = var_4557_end_0, end_mask = var_4557_end_mask_0, squeeze_mask = var_4557_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4557_cast_fp16")]; + bool var_4558_transpose_x_0 = const()[name = string("op_4558_transpose_x_0"), val = bool(false)]; + bool var_4558_transpose_y_0 = const()[name = string("op_4558_transpose_y_0"), val = bool(false)]; + tensor var_4558_cast_fp16 = matmul(transpose_x = var_4558_transpose_x_0, transpose_y = var_4558_transpose_y_0, x = var_4557_cast_fp16, y = state_181_cast_fp16)[name = string("op_4558_cast_fp16")]; + tensor var_4560_begin_0 = const()[name = string("op_4560_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4560_end_0 = const()[name = string("op_4560_end_0"), val = tensor([32, 11, 64, 64])]; + tensor var_4560_end_mask_0 = const()[name = string("op_4560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4560_squeeze_mask_0 = const()[name = string("op_4560_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4560_cast_fp16 = slice_by_index(begin = var_4560_begin_0, end = var_4560_end_0, end_mask = var_4560_end_mask_0, squeeze_mask = var_4560_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4560_cast_fp16")]; + bool var_4561_transpose_x_0 = const()[name = string("op_4561_transpose_x_0"), val = bool(false)]; + bool var_4561_transpose_y_0 = const()[name = string("op_4561_transpose_y_0"), val = bool(false)]; + tensor var_4561_cast_fp16 = matmul(transpose_x = var_4561_transpose_x_0, transpose_y = var_4561_transpose_y_0, x = var_4560_cast_fp16, y = v_new_181_cast_fp16)[name = string("op_4561_cast_fp16")]; + tensor var_4562_cast_fp16 = add(x = var_4558_cast_fp16, y = var_4561_cast_fp16)[name = string("op_4562_cast_fp16")]; + tensor var_4564_begin_0 = const()[name = string("op_4564_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4564_end_0 = const()[name = string("op_4564_end_0"), val = tensor([32, 11, 1, 1])]; + tensor var_4564_end_mask_0 = const()[name = string("op_4564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4564_squeeze_mask_0 = const()[name = string("op_4564_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4564_cast_fp16 = slice_by_index(begin = var_4564_begin_0, end = var_4564_end_0, end_mask = var_4564_end_mask_0, squeeze_mask = var_4564_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4564_cast_fp16")]; + tensor var_4565_cast_fp16 = mul(x = state_181_cast_fp16, y = var_4564_cast_fp16)[name = string("op_4565_cast_fp16")]; + tensor var_4567_begin_0 = const()[name = string("op_4567_begin_0"), val = tensor([0, 10, 0, 0])]; + tensor var_4567_end_0 = const()[name = string("op_4567_end_0"), val = tensor([32, 11, 64, 128])]; + tensor var_4567_end_mask_0 = const()[name = string("op_4567_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4567_squeeze_mask_0 = const()[name = string("op_4567_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4567_cast_fp16 = slice_by_index(begin = var_4567_begin_0, end = var_4567_end_0, end_mask = var_4567_end_mask_0, squeeze_mask = var_4567_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4567_cast_fp16")]; + bool var_4569_transpose_x_1 = const()[name = string("op_4569_transpose_x_1"), val = bool(true)]; + bool var_4569_transpose_y_1 = const()[name = string("op_4569_transpose_y_1"), val = bool(false)]; + tensor var_4569_cast_fp16 = matmul(transpose_x = var_4569_transpose_x_1, transpose_y = var_4569_transpose_y_1, x = var_4567_cast_fp16, y = v_new_181_cast_fp16)[name = string("op_4569_cast_fp16")]; + tensor state_183_cast_fp16 = add(x = var_4565_cast_fp16, y = var_4569_cast_fp16)[name = string("state_183_cast_fp16")]; + tensor var_4572_begin_0 = const()[name = string("op_4572_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4572_end_0 = const()[name = string("op_4572_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_4572_end_mask_0 = const()[name = string("op_4572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4572_squeeze_mask_0 = const()[name = string("op_4572_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4572_cast_fp16 = slice_by_index(begin = var_4572_begin_0, end = var_4572_end_0, end_mask = var_4572_end_mask_0, squeeze_mask = var_4572_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4572_cast_fp16")]; + tensor var_4574_begin_0 = const()[name = string("op_4574_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4574_end_0 = const()[name = string("op_4574_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_4574_end_mask_0 = const()[name = string("op_4574_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4574_squeeze_mask_0 = const()[name = string("op_4574_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4574_cast_fp16 = slice_by_index(begin = var_4574_begin_0, end = var_4574_end_0, end_mask = var_4574_end_mask_0, squeeze_mask = var_4574_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4574_cast_fp16")]; + bool var_4575_transpose_x_0 = const()[name = string("op_4575_transpose_x_0"), val = bool(false)]; + bool var_4575_transpose_y_0 = const()[name = string("op_4575_transpose_y_0"), val = bool(false)]; + tensor var_4575_cast_fp16 = matmul(transpose_x = var_4575_transpose_x_0, transpose_y = var_4575_transpose_y_0, x = var_4574_cast_fp16, y = state_183_cast_fp16)[name = string("op_4575_cast_fp16")]; + tensor v_new_183_cast_fp16 = sub(x = var_4572_cast_fp16, y = var_4575_cast_fp16)[name = string("v_new_183_cast_fp16")]; + tensor var_4578_begin_0 = const()[name = string("op_4578_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4578_end_0 = const()[name = string("op_4578_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_4578_end_mask_0 = const()[name = string("op_4578_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4578_squeeze_mask_0 = const()[name = string("op_4578_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4578_cast_fp16 = slice_by_index(begin = var_4578_begin_0, end = var_4578_end_0, end_mask = var_4578_end_mask_0, squeeze_mask = var_4578_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4578_cast_fp16")]; + bool var_4579_transpose_x_0 = const()[name = string("op_4579_transpose_x_0"), val = bool(false)]; + bool var_4579_transpose_y_0 = const()[name = string("op_4579_transpose_y_0"), val = bool(false)]; + tensor var_4579_cast_fp16 = matmul(transpose_x = var_4579_transpose_x_0, transpose_y = var_4579_transpose_y_0, x = var_4578_cast_fp16, y = state_183_cast_fp16)[name = string("op_4579_cast_fp16")]; + tensor var_4581_begin_0 = const()[name = string("op_4581_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4581_end_0 = const()[name = string("op_4581_end_0"), val = tensor([32, 12, 64, 64])]; + tensor var_4581_end_mask_0 = const()[name = string("op_4581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4581_squeeze_mask_0 = const()[name = string("op_4581_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4581_cast_fp16 = slice_by_index(begin = var_4581_begin_0, end = var_4581_end_0, end_mask = var_4581_end_mask_0, squeeze_mask = var_4581_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4581_cast_fp16")]; + bool var_4582_transpose_x_0 = const()[name = string("op_4582_transpose_x_0"), val = bool(false)]; + bool var_4582_transpose_y_0 = const()[name = string("op_4582_transpose_y_0"), val = bool(false)]; + tensor var_4582_cast_fp16 = matmul(transpose_x = var_4582_transpose_x_0, transpose_y = var_4582_transpose_y_0, x = var_4581_cast_fp16, y = v_new_183_cast_fp16)[name = string("op_4582_cast_fp16")]; + tensor var_4583_cast_fp16 = add(x = var_4579_cast_fp16, y = var_4582_cast_fp16)[name = string("op_4583_cast_fp16")]; + tensor var_4585_begin_0 = const()[name = string("op_4585_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4585_end_0 = const()[name = string("op_4585_end_0"), val = tensor([32, 12, 1, 1])]; + tensor var_4585_end_mask_0 = const()[name = string("op_4585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4585_squeeze_mask_0 = const()[name = string("op_4585_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4585_cast_fp16 = slice_by_index(begin = var_4585_begin_0, end = var_4585_end_0, end_mask = var_4585_end_mask_0, squeeze_mask = var_4585_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4585_cast_fp16")]; + tensor var_4586_cast_fp16 = mul(x = state_183_cast_fp16, y = var_4585_cast_fp16)[name = string("op_4586_cast_fp16")]; + tensor var_4588_begin_0 = const()[name = string("op_4588_begin_0"), val = tensor([0, 11, 0, 0])]; + tensor var_4588_end_0 = const()[name = string("op_4588_end_0"), val = tensor([32, 12, 64, 128])]; + tensor var_4588_end_mask_0 = const()[name = string("op_4588_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4588_squeeze_mask_0 = const()[name = string("op_4588_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4588_cast_fp16 = slice_by_index(begin = var_4588_begin_0, end = var_4588_end_0, end_mask = var_4588_end_mask_0, squeeze_mask = var_4588_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4588_cast_fp16")]; + bool var_4590_transpose_x_1 = const()[name = string("op_4590_transpose_x_1"), val = bool(true)]; + bool var_4590_transpose_y_1 = const()[name = string("op_4590_transpose_y_1"), val = bool(false)]; + tensor var_4590_cast_fp16 = matmul(transpose_x = var_4590_transpose_x_1, transpose_y = var_4590_transpose_y_1, x = var_4588_cast_fp16, y = v_new_183_cast_fp16)[name = string("op_4590_cast_fp16")]; + tensor state_185_cast_fp16 = add(x = var_4586_cast_fp16, y = var_4590_cast_fp16)[name = string("state_185_cast_fp16")]; + tensor var_4593_begin_0 = const()[name = string("op_4593_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4593_end_0 = const()[name = string("op_4593_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_4593_end_mask_0 = const()[name = string("op_4593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4593_squeeze_mask_0 = const()[name = string("op_4593_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4593_cast_fp16 = slice_by_index(begin = var_4593_begin_0, end = var_4593_end_0, end_mask = var_4593_end_mask_0, squeeze_mask = var_4593_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4593_cast_fp16")]; + tensor var_4595_begin_0 = const()[name = string("op_4595_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4595_end_0 = const()[name = string("op_4595_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_4595_end_mask_0 = const()[name = string("op_4595_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4595_squeeze_mask_0 = const()[name = string("op_4595_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4595_cast_fp16 = slice_by_index(begin = var_4595_begin_0, end = var_4595_end_0, end_mask = var_4595_end_mask_0, squeeze_mask = var_4595_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4595_cast_fp16")]; + bool var_4596_transpose_x_0 = const()[name = string("op_4596_transpose_x_0"), val = bool(false)]; + bool var_4596_transpose_y_0 = const()[name = string("op_4596_transpose_y_0"), val = bool(false)]; + tensor var_4596_cast_fp16 = matmul(transpose_x = var_4596_transpose_x_0, transpose_y = var_4596_transpose_y_0, x = var_4595_cast_fp16, y = state_185_cast_fp16)[name = string("op_4596_cast_fp16")]; + tensor v_new_185_cast_fp16 = sub(x = var_4593_cast_fp16, y = var_4596_cast_fp16)[name = string("v_new_185_cast_fp16")]; + tensor var_4599_begin_0 = const()[name = string("op_4599_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4599_end_0 = const()[name = string("op_4599_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_4599_end_mask_0 = const()[name = string("op_4599_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4599_squeeze_mask_0 = const()[name = string("op_4599_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4599_cast_fp16 = slice_by_index(begin = var_4599_begin_0, end = var_4599_end_0, end_mask = var_4599_end_mask_0, squeeze_mask = var_4599_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4599_cast_fp16")]; + bool var_4600_transpose_x_0 = const()[name = string("op_4600_transpose_x_0"), val = bool(false)]; + bool var_4600_transpose_y_0 = const()[name = string("op_4600_transpose_y_0"), val = bool(false)]; + tensor var_4600_cast_fp16 = matmul(transpose_x = var_4600_transpose_x_0, transpose_y = var_4600_transpose_y_0, x = var_4599_cast_fp16, y = state_185_cast_fp16)[name = string("op_4600_cast_fp16")]; + tensor var_4602_begin_0 = const()[name = string("op_4602_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4602_end_0 = const()[name = string("op_4602_end_0"), val = tensor([32, 13, 64, 64])]; + tensor var_4602_end_mask_0 = const()[name = string("op_4602_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4602_squeeze_mask_0 = const()[name = string("op_4602_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4602_cast_fp16 = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, squeeze_mask = var_4602_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4602_cast_fp16")]; + bool var_4603_transpose_x_0 = const()[name = string("op_4603_transpose_x_0"), val = bool(false)]; + bool var_4603_transpose_y_0 = const()[name = string("op_4603_transpose_y_0"), val = bool(false)]; + tensor var_4603_cast_fp16 = matmul(transpose_x = var_4603_transpose_x_0, transpose_y = var_4603_transpose_y_0, x = var_4602_cast_fp16, y = v_new_185_cast_fp16)[name = string("op_4603_cast_fp16")]; + tensor var_4604_cast_fp16 = add(x = var_4600_cast_fp16, y = var_4603_cast_fp16)[name = string("op_4604_cast_fp16")]; + tensor var_4606_begin_0 = const()[name = string("op_4606_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4606_end_0 = const()[name = string("op_4606_end_0"), val = tensor([32, 13, 1, 1])]; + tensor var_4606_end_mask_0 = const()[name = string("op_4606_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4606_squeeze_mask_0 = const()[name = string("op_4606_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4606_cast_fp16 = slice_by_index(begin = var_4606_begin_0, end = var_4606_end_0, end_mask = var_4606_end_mask_0, squeeze_mask = var_4606_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4606_cast_fp16")]; + tensor var_4607_cast_fp16 = mul(x = state_185_cast_fp16, y = var_4606_cast_fp16)[name = string("op_4607_cast_fp16")]; + tensor var_4609_begin_0 = const()[name = string("op_4609_begin_0"), val = tensor([0, 12, 0, 0])]; + tensor var_4609_end_0 = const()[name = string("op_4609_end_0"), val = tensor([32, 13, 64, 128])]; + tensor var_4609_end_mask_0 = const()[name = string("op_4609_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4609_squeeze_mask_0 = const()[name = string("op_4609_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4609_cast_fp16 = slice_by_index(begin = var_4609_begin_0, end = var_4609_end_0, end_mask = var_4609_end_mask_0, squeeze_mask = var_4609_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4609_cast_fp16")]; + bool var_4611_transpose_x_1 = const()[name = string("op_4611_transpose_x_1"), val = bool(true)]; + bool var_4611_transpose_y_1 = const()[name = string("op_4611_transpose_y_1"), val = bool(false)]; + tensor var_4611_cast_fp16 = matmul(transpose_x = var_4611_transpose_x_1, transpose_y = var_4611_transpose_y_1, x = var_4609_cast_fp16, y = v_new_185_cast_fp16)[name = string("op_4611_cast_fp16")]; + tensor state_187_cast_fp16 = add(x = var_4607_cast_fp16, y = var_4611_cast_fp16)[name = string("state_187_cast_fp16")]; + tensor var_4614_begin_0 = const()[name = string("op_4614_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4614_end_0 = const()[name = string("op_4614_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_4614_end_mask_0 = const()[name = string("op_4614_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4614_squeeze_mask_0 = const()[name = string("op_4614_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4614_cast_fp16 = slice_by_index(begin = var_4614_begin_0, end = var_4614_end_0, end_mask = var_4614_end_mask_0, squeeze_mask = var_4614_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4614_cast_fp16")]; + tensor var_4616_begin_0 = const()[name = string("op_4616_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4616_end_0 = const()[name = string("op_4616_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_4616_end_mask_0 = const()[name = string("op_4616_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4616_squeeze_mask_0 = const()[name = string("op_4616_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4616_cast_fp16 = slice_by_index(begin = var_4616_begin_0, end = var_4616_end_0, end_mask = var_4616_end_mask_0, squeeze_mask = var_4616_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4616_cast_fp16")]; + bool var_4617_transpose_x_0 = const()[name = string("op_4617_transpose_x_0"), val = bool(false)]; + bool var_4617_transpose_y_0 = const()[name = string("op_4617_transpose_y_0"), val = bool(false)]; + tensor var_4617_cast_fp16 = matmul(transpose_x = var_4617_transpose_x_0, transpose_y = var_4617_transpose_y_0, x = var_4616_cast_fp16, y = state_187_cast_fp16)[name = string("op_4617_cast_fp16")]; + tensor v_new_187_cast_fp16 = sub(x = var_4614_cast_fp16, y = var_4617_cast_fp16)[name = string("v_new_187_cast_fp16")]; + tensor var_4620_begin_0 = const()[name = string("op_4620_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4620_end_0 = const()[name = string("op_4620_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_4620_end_mask_0 = const()[name = string("op_4620_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4620_squeeze_mask_0 = const()[name = string("op_4620_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4620_cast_fp16 = slice_by_index(begin = var_4620_begin_0, end = var_4620_end_0, end_mask = var_4620_end_mask_0, squeeze_mask = var_4620_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4620_cast_fp16")]; + bool var_4621_transpose_x_0 = const()[name = string("op_4621_transpose_x_0"), val = bool(false)]; + bool var_4621_transpose_y_0 = const()[name = string("op_4621_transpose_y_0"), val = bool(false)]; + tensor var_4621_cast_fp16 = matmul(transpose_x = var_4621_transpose_x_0, transpose_y = var_4621_transpose_y_0, x = var_4620_cast_fp16, y = state_187_cast_fp16)[name = string("op_4621_cast_fp16")]; + tensor var_4623_begin_0 = const()[name = string("op_4623_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4623_end_0 = const()[name = string("op_4623_end_0"), val = tensor([32, 14, 64, 64])]; + tensor var_4623_end_mask_0 = const()[name = string("op_4623_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4623_squeeze_mask_0 = const()[name = string("op_4623_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4623_cast_fp16 = slice_by_index(begin = var_4623_begin_0, end = var_4623_end_0, end_mask = var_4623_end_mask_0, squeeze_mask = var_4623_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4623_cast_fp16")]; + bool var_4624_transpose_x_0 = const()[name = string("op_4624_transpose_x_0"), val = bool(false)]; + bool var_4624_transpose_y_0 = const()[name = string("op_4624_transpose_y_0"), val = bool(false)]; + tensor var_4624_cast_fp16 = matmul(transpose_x = var_4624_transpose_x_0, transpose_y = var_4624_transpose_y_0, x = var_4623_cast_fp16, y = v_new_187_cast_fp16)[name = string("op_4624_cast_fp16")]; + tensor var_4625_cast_fp16 = add(x = var_4621_cast_fp16, y = var_4624_cast_fp16)[name = string("op_4625_cast_fp16")]; + tensor var_4627_begin_0 = const()[name = string("op_4627_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4627_end_0 = const()[name = string("op_4627_end_0"), val = tensor([32, 14, 1, 1])]; + tensor var_4627_end_mask_0 = const()[name = string("op_4627_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4627_squeeze_mask_0 = const()[name = string("op_4627_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4627_cast_fp16 = slice_by_index(begin = var_4627_begin_0, end = var_4627_end_0, end_mask = var_4627_end_mask_0, squeeze_mask = var_4627_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4627_cast_fp16")]; + tensor var_4628_cast_fp16 = mul(x = state_187_cast_fp16, y = var_4627_cast_fp16)[name = string("op_4628_cast_fp16")]; + tensor var_4630_begin_0 = const()[name = string("op_4630_begin_0"), val = tensor([0, 13, 0, 0])]; + tensor var_4630_end_0 = const()[name = string("op_4630_end_0"), val = tensor([32, 14, 64, 128])]; + tensor var_4630_end_mask_0 = const()[name = string("op_4630_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4630_squeeze_mask_0 = const()[name = string("op_4630_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4630_cast_fp16 = slice_by_index(begin = var_4630_begin_0, end = var_4630_end_0, end_mask = var_4630_end_mask_0, squeeze_mask = var_4630_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4630_cast_fp16")]; + bool var_4632_transpose_x_1 = const()[name = string("op_4632_transpose_x_1"), val = bool(true)]; + bool var_4632_transpose_y_1 = const()[name = string("op_4632_transpose_y_1"), val = bool(false)]; + tensor var_4632_cast_fp16 = matmul(transpose_x = var_4632_transpose_x_1, transpose_y = var_4632_transpose_y_1, x = var_4630_cast_fp16, y = v_new_187_cast_fp16)[name = string("op_4632_cast_fp16")]; + tensor state_189_cast_fp16 = add(x = var_4628_cast_fp16, y = var_4632_cast_fp16)[name = string("state_189_cast_fp16")]; + tensor var_4635_begin_0 = const()[name = string("op_4635_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4635_end_0 = const()[name = string("op_4635_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_4635_end_mask_0 = const()[name = string("op_4635_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4635_squeeze_mask_0 = const()[name = string("op_4635_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4635_cast_fp16 = slice_by_index(begin = var_4635_begin_0, end = var_4635_end_0, end_mask = var_4635_end_mask_0, squeeze_mask = var_4635_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4635_cast_fp16")]; + tensor var_4637_begin_0 = const()[name = string("op_4637_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4637_end_0 = const()[name = string("op_4637_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_4637_end_mask_0 = const()[name = string("op_4637_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4637_squeeze_mask_0 = const()[name = string("op_4637_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4637_cast_fp16 = slice_by_index(begin = var_4637_begin_0, end = var_4637_end_0, end_mask = var_4637_end_mask_0, squeeze_mask = var_4637_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4637_cast_fp16")]; + bool var_4638_transpose_x_0 = const()[name = string("op_4638_transpose_x_0"), val = bool(false)]; + bool var_4638_transpose_y_0 = const()[name = string("op_4638_transpose_y_0"), val = bool(false)]; + tensor var_4638_cast_fp16 = matmul(transpose_x = var_4638_transpose_x_0, transpose_y = var_4638_transpose_y_0, x = var_4637_cast_fp16, y = state_189_cast_fp16)[name = string("op_4638_cast_fp16")]; + tensor v_new_189_cast_fp16 = sub(x = var_4635_cast_fp16, y = var_4638_cast_fp16)[name = string("v_new_189_cast_fp16")]; + tensor var_4641_begin_0 = const()[name = string("op_4641_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4641_end_0 = const()[name = string("op_4641_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_4641_end_mask_0 = const()[name = string("op_4641_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4641_squeeze_mask_0 = const()[name = string("op_4641_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4641_cast_fp16 = slice_by_index(begin = var_4641_begin_0, end = var_4641_end_0, end_mask = var_4641_end_mask_0, squeeze_mask = var_4641_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4641_cast_fp16")]; + bool var_4642_transpose_x_0 = const()[name = string("op_4642_transpose_x_0"), val = bool(false)]; + bool var_4642_transpose_y_0 = const()[name = string("op_4642_transpose_y_0"), val = bool(false)]; + tensor var_4642_cast_fp16 = matmul(transpose_x = var_4642_transpose_x_0, transpose_y = var_4642_transpose_y_0, x = var_4641_cast_fp16, y = state_189_cast_fp16)[name = string("op_4642_cast_fp16")]; + tensor var_4644_begin_0 = const()[name = string("op_4644_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4644_end_0 = const()[name = string("op_4644_end_0"), val = tensor([32, 15, 64, 64])]; + tensor var_4644_end_mask_0 = const()[name = string("op_4644_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4644_squeeze_mask_0 = const()[name = string("op_4644_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4644_cast_fp16 = slice_by_index(begin = var_4644_begin_0, end = var_4644_end_0, end_mask = var_4644_end_mask_0, squeeze_mask = var_4644_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4644_cast_fp16")]; + bool var_4645_transpose_x_0 = const()[name = string("op_4645_transpose_x_0"), val = bool(false)]; + bool var_4645_transpose_y_0 = const()[name = string("op_4645_transpose_y_0"), val = bool(false)]; + tensor var_4645_cast_fp16 = matmul(transpose_x = var_4645_transpose_x_0, transpose_y = var_4645_transpose_y_0, x = var_4644_cast_fp16, y = v_new_189_cast_fp16)[name = string("op_4645_cast_fp16")]; + tensor var_4646_cast_fp16 = add(x = var_4642_cast_fp16, y = var_4645_cast_fp16)[name = string("op_4646_cast_fp16")]; + tensor var_4648_begin_0 = const()[name = string("op_4648_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4648_end_0 = const()[name = string("op_4648_end_0"), val = tensor([32, 15, 1, 1])]; + tensor var_4648_end_mask_0 = const()[name = string("op_4648_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4648_squeeze_mask_0 = const()[name = string("op_4648_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4648_cast_fp16 = slice_by_index(begin = var_4648_begin_0, end = var_4648_end_0, end_mask = var_4648_end_mask_0, squeeze_mask = var_4648_squeeze_mask_0, x = chunk_decay_cast_fp16)[name = string("op_4648_cast_fp16")]; + tensor var_4649_cast_fp16 = mul(x = state_189_cast_fp16, y = var_4648_cast_fp16)[name = string("op_4649_cast_fp16")]; + tensor var_4651_begin_0 = const()[name = string("op_4651_begin_0"), val = tensor([0, 14, 0, 0])]; + tensor var_4651_end_0 = const()[name = string("op_4651_end_0"), val = tensor([32, 15, 64, 128])]; + tensor var_4651_end_mask_0 = const()[name = string("op_4651_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4651_squeeze_mask_0 = const()[name = string("op_4651_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4651_cast_fp16 = slice_by_index(begin = var_4651_begin_0, end = var_4651_end_0, end_mask = var_4651_end_mask_0, squeeze_mask = var_4651_squeeze_mask_0, x = k_dec_cast_fp16)[name = string("op_4651_cast_fp16")]; + bool var_4653_transpose_x_1 = const()[name = string("op_4653_transpose_x_1"), val = bool(true)]; + bool var_4653_transpose_y_1 = const()[name = string("op_4653_transpose_y_1"), val = bool(false)]; + tensor var_4653_cast_fp16 = matmul(transpose_x = var_4653_transpose_x_1, transpose_y = var_4653_transpose_y_1, x = var_4651_cast_fp16, y = v_new_189_cast_fp16)[name = string("op_4653_cast_fp16")]; + tensor state_cast_fp16 = add(x = var_4649_cast_fp16, y = var_4653_cast_fp16)[name = string("state_cast_fp16")]; + tensor var_4656_begin_0 = const()[name = string("op_4656_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_4656_end_0 = const()[name = string("op_4656_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_4656_end_mask_0 = const()[name = string("op_4656_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4656_squeeze_mask_0 = const()[name = string("op_4656_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4656_cast_fp16 = slice_by_index(begin = var_4656_begin_0, end = var_4656_end_0, end_mask = var_4656_end_mask_0, squeeze_mask = var_4656_squeeze_mask_0, x = new_v_cast_fp16)[name = string("op_4656_cast_fp16")]; + tensor var_4658_begin_0 = const()[name = string("op_4658_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_4658_end_0 = const()[name = string("op_4658_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_4658_end_mask_0 = const()[name = string("op_4658_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4658_squeeze_mask_0 = const()[name = string("op_4658_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4658_cast_fp16 = slice_by_index(begin = var_4658_begin_0, end = var_4658_end_0, end_mask = var_4658_end_mask_0, squeeze_mask = var_4658_squeeze_mask_0, x = k_cumdecay_cast_fp16)[name = string("op_4658_cast_fp16")]; + bool var_4659_transpose_x_0 = const()[name = string("op_4659_transpose_x_0"), val = bool(false)]; + bool var_4659_transpose_y_0 = const()[name = string("op_4659_transpose_y_0"), val = bool(false)]; + tensor var_4659_cast_fp16 = matmul(transpose_x = var_4659_transpose_x_0, transpose_y = var_4659_transpose_y_0, x = var_4658_cast_fp16, y = state_cast_fp16)[name = string("op_4659_cast_fp16")]; + tensor v_new_cast_fp16 = sub(x = var_4656_cast_fp16, y = var_4659_cast_fp16)[name = string("v_new_cast_fp16")]; + tensor var_4662_begin_0 = const()[name = string("op_4662_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_4662_end_0 = const()[name = string("op_4662_end_0"), val = tensor([32, 16, 64, 128])]; + tensor var_4662_end_mask_0 = const()[name = string("op_4662_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4662_squeeze_mask_0 = const()[name = string("op_4662_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4662_cast_fp16 = slice_by_index(begin = var_4662_begin_0, end = var_4662_end_0, end_mask = var_4662_end_mask_0, squeeze_mask = var_4662_squeeze_mask_0, x = q_dec_cast_fp16)[name = string("op_4662_cast_fp16")]; + bool var_4663_transpose_x_0 = const()[name = string("op_4663_transpose_x_0"), val = bool(false)]; + bool var_4663_transpose_y_0 = const()[name = string("op_4663_transpose_y_0"), val = bool(false)]; + tensor var_4663_cast_fp16 = matmul(transpose_x = var_4663_transpose_x_0, transpose_y = var_4663_transpose_y_0, x = var_4662_cast_fp16, y = state_cast_fp16)[name = string("op_4663_cast_fp16")]; + tensor var_4665_begin_0 = const()[name = string("op_4665_begin_0"), val = tensor([0, 15, 0, 0])]; + tensor var_4665_end_0 = const()[name = string("op_4665_end_0"), val = tensor([32, 16, 64, 64])]; + tensor var_4665_end_mask_0 = const()[name = string("op_4665_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_4665_squeeze_mask_0 = const()[name = string("op_4665_squeeze_mask_0"), val = tensor([false, true, false, false])]; + tensor var_4665_cast_fp16 = slice_by_index(begin = var_4665_begin_0, end = var_4665_end_0, end_mask = var_4665_end_mask_0, squeeze_mask = var_4665_squeeze_mask_0, x = attn_intra_cast_fp16)[name = string("op_4665_cast_fp16")]; + bool var_4666_transpose_x_0 = const()[name = string("op_4666_transpose_x_0"), val = bool(false)]; + bool var_4666_transpose_y_0 = const()[name = string("op_4666_transpose_y_0"), val = bool(false)]; + tensor var_4666_cast_fp16 = matmul(transpose_x = var_4666_transpose_x_0, transpose_y = var_4666_transpose_y_0, x = var_4665_cast_fp16, y = v_new_cast_fp16)[name = string("op_4666_cast_fp16")]; + tensor var_4667_cast_fp16 = add(x = var_4663_cast_fp16, y = var_4666_cast_fp16)[name = string("op_4667_cast_fp16")]; + int32 var_4669_axis_0 = const()[name = string("op_4669_axis_0"), val = int32(1)]; + tensor var_4669_cast_fp16 = stack(axis = var_4669_axis_0, values = (var_4352_cast_fp16, var_4373_cast_fp16, var_4394_cast_fp16, var_4415_cast_fp16, var_4436_cast_fp16, var_4457_cast_fp16, var_4478_cast_fp16, var_4499_cast_fp16, var_4520_cast_fp16, var_4541_cast_fp16, var_4562_cast_fp16, var_4583_cast_fp16, var_4604_cast_fp16, var_4625_cast_fp16, var_4646_cast_fp16, var_4667_cast_fp16))[name = string("op_4669_cast_fp16")]; + tensor var_4670 = const()[name = string("op_4670"), val = tensor([32, 1024, 128])]; + tensor var_4671_cast_fp16 = reshape(shape = var_4670, x = var_4669_cast_fp16)[name = string("op_4671_cast_fp16")]; + tensor var_4672 = const()[name = string("op_4672"), val = tensor([1, 0, 2])]; + tensor input_95_begin_0 = const()[name = string("input_95_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor input_95_end_0 = const()[name = string("input_95_end_0"), val = tensor([1, 1024, 32, 128])]; + tensor input_95_end_mask_0 = const()[name = string("input_95_end_mask_0"), val = tensor([false, true, true, true])]; + tensor input_95_squeeze_mask_0 = const()[name = string("input_95_squeeze_mask_0"), val = tensor([true, false, false, false])]; + tensor input_95_cast_fp16 = slice_by_index(begin = input_95_begin_0, end = input_95_end_0, end_mask = input_95_end_mask_0, squeeze_mask = input_95_squeeze_mask_0, x = z_cast_fp16)[name = string("input_95_cast_fp16")]; + fp16 var_3994_promoted_1_to_fp16 = const()[name = string("op_3994_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor x_169_cast_fp16 = transpose(perm = var_4672, x = var_4671_cast_fp16)[name = string("transpose_4")]; + tensor var_4676_cast_fp16 = pow(x = x_169_cast_fp16, y = var_3994_promoted_1_to_fp16)[name = string("op_4676_cast_fp16")]; + tensor var_4678_axes_0 = const()[name = string("op_4678_axes_0"), val = tensor([-1])]; + bool var_4678_keep_dims_0 = const()[name = string("op_4678_keep_dims_0"), val = bool(true)]; + tensor var_4678_cast_fp16 = reduce_mean(axes = var_4678_axes_0, keep_dims = var_4678_keep_dims_0, x = var_4676_cast_fp16)[name = string("op_4678_cast_fp16")]; + fp16 var_4679_to_fp16 = const()[name = string("op_4679_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4680_cast_fp16 = add(x = var_4678_cast_fp16, y = var_4679_to_fp16)[name = string("op_4680_cast_fp16")]; + fp32 var_4681_epsilon_0 = const()[name = string("op_4681_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4681_cast_fp16 = rsqrt(epsilon = var_4681_epsilon_0, x = var_4680_cast_fp16)[name = string("op_4681_cast_fp16")]; + tensor x_171_cast_fp16 = mul(x = x_169_cast_fp16, y = var_4681_cast_fp16)[name = string("x_171_cast_fp16")]; + tensor layers_6_linear_attn_norm_weight_to_fp16 = const()[name = string("layers_6_linear_attn_norm_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413416000)))]; + tensor var_4683_cast_fp16 = mul(x = layers_6_linear_attn_norm_weight_to_fp16, y = x_171_cast_fp16)[name = string("op_4683_cast_fp16")]; + tensor var_4684_cast_fp16 = silu(x = input_95_cast_fp16)[name = string("op_4684_cast_fp16")]; + tensor core_cast_fp16 = mul(x = var_4683_cast_fp16, y = var_4684_cast_fp16)[name = string("core_cast_fp16")]; + tensor var_4686 = const()[name = string("op_4686"), val = tensor([1, 1024, 4096])]; + tensor input_97_cast_fp16 = reshape(shape = var_4686, x = core_cast_fp16)[name = string("input_97_cast_fp16")]; + tensor layers_6_linear_attn_out_proj_weight_to_fp16 = const()[name = string("layers_6_linear_attn_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1413416320)))]; + tensor linear_51_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_6_linear_attn_out_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("linear_51_cast_fp16")]; + tensor x_173_cast_fp16 = add(x = x_151_cast_fp16, y = linear_51_cast_fp16)[name = string("x_173_cast_fp16")]; + fp16 var_3994_promoted_2_to_fp16 = const()[name = string("op_3994_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4692_cast_fp16 = pow(x = x_173_cast_fp16, y = var_3994_promoted_2_to_fp16)[name = string("op_4692_cast_fp16")]; + tensor var_4694_axes_0 = const()[name = string("op_4694_axes_0"), val = tensor([-1])]; + bool var_4694_keep_dims_0 = const()[name = string("op_4694_keep_dims_0"), val = bool(true)]; + tensor var_4694_cast_fp16 = reduce_mean(axes = var_4694_axes_0, keep_dims = var_4694_keep_dims_0, x = var_4692_cast_fp16)[name = string("op_4694_cast_fp16")]; + fp16 var_4695_to_fp16 = const()[name = string("op_4695_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4696_cast_fp16 = add(x = var_4694_cast_fp16, y = var_4695_to_fp16)[name = string("op_4696_cast_fp16")]; + fp32 var_4697_epsilon_0 = const()[name = string("op_4697_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4697_cast_fp16 = rsqrt(epsilon = var_4697_epsilon_0, x = var_4696_cast_fp16)[name = string("op_4697_cast_fp16")]; + tensor x_175_cast_fp16 = mul(x = x_173_cast_fp16, y = var_4697_cast_fp16)[name = string("x_175_cast_fp16")]; + tensor var_4700_to_fp16 = const()[name = string("op_4700_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1434387904)))]; + tensor input_99_cast_fp16 = mul(x = x_175_cast_fp16, y = var_4700_to_fp16)[name = string("input_99_cast_fp16")]; + tensor layers_6_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1434393088)))]; + tensor linear_52_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_6_mlp_gate_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_52_cast_fp16")]; + tensor var_4707_cast_fp16 = silu(x = linear_52_cast_fp16)[name = string("op_4707_cast_fp16")]; + tensor layers_6_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1481579072)))]; + tensor linear_53_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_6_mlp_up_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_53_cast_fp16")]; + tensor input_103_cast_fp16 = mul(x = var_4707_cast_fp16, y = linear_53_cast_fp16)[name = string("input_103_cast_fp16")]; + tensor layers_6_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1528765056)))]; + tensor linear_54_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_6_mlp_down_proj_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_54_cast_fp16")]; + tensor x_177_cast_fp16 = add(x = x_173_cast_fp16, y = linear_54_cast_fp16)[name = string("x_177_cast_fp16")]; + int32 var_4727 = const()[name = string("op_4727"), val = int32(-1)]; + fp16 var_4726_promoted_to_fp16 = const()[name = string("op_4726_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4736_cast_fp16 = pow(x = x_177_cast_fp16, y = var_4726_promoted_to_fp16)[name = string("op_4736_cast_fp16")]; + tensor var_4738_axes_0 = const()[name = string("op_4738_axes_0"), val = tensor([-1])]; + bool var_4738_keep_dims_0 = const()[name = string("op_4738_keep_dims_0"), val = bool(true)]; + tensor var_4738_cast_fp16 = reduce_mean(axes = var_4738_axes_0, keep_dims = var_4738_keep_dims_0, x = var_4736_cast_fp16)[name = string("op_4738_cast_fp16")]; + fp16 var_4739_to_fp16 = const()[name = string("op_4739_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4740_cast_fp16 = add(x = var_4738_cast_fp16, y = var_4739_to_fp16)[name = string("op_4740_cast_fp16")]; + fp32 var_4741_epsilon_0 = const()[name = string("op_4741_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4741_cast_fp16 = rsqrt(epsilon = var_4741_epsilon_0, x = var_4740_cast_fp16)[name = string("op_4741_cast_fp16")]; + tensor x_179_cast_fp16 = mul(x = x_177_cast_fp16, y = var_4741_cast_fp16)[name = string("x_179_cast_fp16")]; + tensor var_4744_to_fp16 = const()[name = string("op_4744_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1575951040)))]; + tensor x_181_cast_fp16 = mul(x = x_179_cast_fp16, y = var_4744_to_fp16)[name = string("x_181_cast_fp16")]; + tensor layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1575956224)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = x_181_cast_fp16)[name = string("linear_55_cast_fp16")]; + tensor var_4755 = const()[name = string("op_4755"), val = tensor([1, 1024, 16, 512])]; + tensor qg_cast_fp16 = reshape(shape = var_4755, x = linear_55_cast_fp16)[name = string("qg_cast_fp16")]; + tensor x_183_begin_0 = const()[name = string("x_183_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor x_183_end_0 = const()[name = string("x_183_end_0"), val = tensor([1, 1024, 16, 256])]; + tensor x_183_end_mask_0 = const()[name = string("x_183_end_mask_0"), val = tensor([true, true, true, false])]; + tensor x_183_cast_fp16 = slice_by_index(begin = x_183_begin_0, end = x_183_end_0, end_mask = x_183_end_mask_0, x = qg_cast_fp16)[name = string("x_183_cast_fp16")]; + tensor gate_5_begin_0 = const()[name = string("gate_5_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor gate_5_end_0 = const()[name = string("gate_5_end_0"), val = tensor([1, 1024, 16, 512])]; + tensor gate_5_end_mask_0 = const()[name = string("gate_5_end_mask_0"), val = tensor([true, true, true, true])]; + tensor gate_5_cast_fp16 = slice_by_index(begin = gate_5_begin_0, end = gate_5_end_0, end_mask = gate_5_end_mask_0, x = qg_cast_fp16)[name = string("gate_5_cast_fp16")]; + tensor var_4759 = const()[name = string("op_4759"), val = tensor([1, 1024, 4096])]; + tensor gate_cast_fp16 = reshape(shape = var_4759, x = gate_5_cast_fp16)[name = string("gate_cast_fp16")]; + fp16 var_4726_promoted_1_to_fp16 = const()[name = string("op_4726_promoted_1_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4762_cast_fp16 = pow(x = x_183_cast_fp16, y = var_4726_promoted_1_to_fp16)[name = string("op_4762_cast_fp16")]; + tensor var_4764_axes_0 = const()[name = string("op_4764_axes_0"), val = tensor([-1])]; + bool var_4764_keep_dims_0 = const()[name = string("op_4764_keep_dims_0"), val = bool(true)]; + tensor var_4764_cast_fp16 = reduce_mean(axes = var_4764_axes_0, keep_dims = var_4764_keep_dims_0, x = var_4762_cast_fp16)[name = string("op_4764_cast_fp16")]; + fp16 var_4765_to_fp16 = const()[name = string("op_4765_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4766_cast_fp16 = add(x = var_4764_cast_fp16, y = var_4765_to_fp16)[name = string("op_4766_cast_fp16")]; + fp32 var_4767_epsilon_0 = const()[name = string("op_4767_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4767_cast_fp16 = rsqrt(epsilon = var_4767_epsilon_0, x = var_4766_cast_fp16)[name = string("op_4767_cast_fp16")]; + tensor x_185_cast_fp16 = mul(x = x_183_cast_fp16, y = var_4767_cast_fp16)[name = string("x_185_cast_fp16")]; + tensor var_4770_to_fp16 = const()[name = string("op_4770_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1617899328)))]; + tensor var_4771_cast_fp16 = mul(x = x_185_cast_fp16, y = var_4770_to_fp16)[name = string("op_4771_cast_fp16")]; + tensor q_65_perm_0 = const()[name = string("q_65_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1617899904)))]; + tensor linear_56_cast_fp16 = linear(bias = linear_25_bias_0_to_fp16, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = x_181_cast_fp16)[name = string("linear_56_cast_fp16")]; + tensor var_4775 = const()[name = string("op_4775"), val = tensor([1, 1024, 4, 256])]; + tensor x_187_cast_fp16 = reshape(shape = var_4775, x = linear_56_cast_fp16)[name = string("x_187_cast_fp16")]; + fp16 var_4726_promoted_2_to_fp16 = const()[name = string("op_4726_promoted_2_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4778_cast_fp16 = pow(x = x_187_cast_fp16, y = var_4726_promoted_2_to_fp16)[name = string("op_4778_cast_fp16")]; + tensor var_4780_axes_0 = const()[name = string("op_4780_axes_0"), val = tensor([-1])]; + bool var_4780_keep_dims_0 = const()[name = string("op_4780_keep_dims_0"), val = bool(true)]; + tensor var_4780_cast_fp16 = reduce_mean(axes = var_4780_axes_0, keep_dims = var_4780_keep_dims_0, x = var_4778_cast_fp16)[name = string("op_4780_cast_fp16")]; + fp16 var_4781_to_fp16 = const()[name = string("op_4781_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4782_cast_fp16 = add(x = var_4780_cast_fp16, y = var_4781_to_fp16)[name = string("op_4782_cast_fp16")]; + fp32 var_4783_epsilon_0 = const()[name = string("op_4783_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4783_cast_fp16 = rsqrt(epsilon = var_4783_epsilon_0, x = var_4782_cast_fp16)[name = string("op_4783_cast_fp16")]; + tensor x_189_cast_fp16 = mul(x = x_187_cast_fp16, y = var_4783_cast_fp16)[name = string("x_189_cast_fp16")]; + tensor var_4786_to_fp16 = const()[name = string("op_4786_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1623142848)))]; + tensor var_4787_cast_fp16 = mul(x = x_189_cast_fp16, y = var_4786_to_fp16)[name = string("op_4787_cast_fp16")]; + tensor k_55_perm_0 = const()[name = string("k_55_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1623143424)))]; + tensor linear_57_cast_fp16 = linear(bias = linear_25_bias_0_to_fp16, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = x_181_cast_fp16)[name = string("linear_57_cast_fp16")]; + tensor var_4791 = const()[name = string("op_4791"), val = tensor([1, 1024, 4, 256])]; + tensor var_4792_cast_fp16 = reshape(shape = var_4791, x = linear_57_cast_fp16)[name = string("op_4792_cast_fp16")]; + tensor v_29_perm_0 = const()[name = string("v_29_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_4798_begin_0 = const()[name = string("op_4798_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4798_end_0 = const()[name = string("op_4798_end_0"), val = tensor([1, 16, 1024, 64])]; + tensor var_4798_end_mask_0 = const()[name = string("op_4798_end_mask_0"), val = tensor([true, true, true, false])]; + tensor q_65_cast_fp16 = transpose(perm = q_65_perm_0, x = var_4771_cast_fp16)[name = string("transpose_3")]; + tensor var_4798_cast_fp16 = slice_by_index(begin = var_4798_begin_0, end = var_4798_end_0, end_mask = var_4798_end_mask_0, x = q_65_cast_fp16)[name = string("op_4798_cast_fp16")]; + tensor var_4799_cast_fp16 = mul(x = var_4798_cast_fp16, y = c_1_cast_fp16)[name = string("op_4799_cast_fp16")]; + tensor var_4806_begin_0 = const()[name = string("op_4806_begin_0"), val = tensor([0, 0, 0, 32])]; + tensor var_4806_end_0 = const()[name = string("op_4806_end_0"), val = tensor([1, 16, 1024, 64])]; + tensor var_4806_end_mask_0 = const()[name = string("op_4806_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4806_cast_fp16 = slice_by_index(begin = var_4806_begin_0, end = var_4806_end_0, end_mask = var_4806_end_mask_0, x = var_4798_cast_fp16)[name = string("op_4806_cast_fp16")]; + fp16 const_79_promoted_to_fp16 = const()[name = string("const_79_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_4807_cast_fp16 = mul(x = var_4806_cast_fp16, y = const_79_promoted_to_fp16)[name = string("op_4807_cast_fp16")]; + tensor var_4808_begin_0 = const()[name = string("op_4808_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4808_end_0 = const()[name = string("op_4808_end_0"), val = tensor([1, 16, 1024, 32])]; + tensor var_4808_end_mask_0 = const()[name = string("op_4808_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4808_cast_fp16 = slice_by_index(begin = var_4808_begin_0, end = var_4808_end_0, end_mask = var_4808_end_mask_0, x = var_4798_cast_fp16)[name = string("op_4808_cast_fp16")]; + bool var_4810_interleave_0 = const()[name = string("op_4810_interleave_0"), val = bool(false)]; + tensor var_4810_cast_fp16 = concat(axis = var_4727, interleave = var_4810_interleave_0, values = (var_4807_cast_fp16, var_4808_cast_fp16))[name = string("op_4810_cast_fp16")]; + tensor var_4811_cast_fp16 = mul(x = var_4810_cast_fp16, y = s_1_cast_fp16)[name = string("op_4811_cast_fp16")]; + tensor var_4812_cast_fp16 = add(x = var_4799_cast_fp16, y = var_4811_cast_fp16)[name = string("op_4812_cast_fp16")]; + tensor var_4813_begin_0 = const()[name = string("op_4813_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4813_end_0 = const()[name = string("op_4813_end_0"), val = tensor([1, 16, 1024, 256])]; + tensor var_4813_end_mask_0 = const()[name = string("op_4813_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4813_cast_fp16 = slice_by_index(begin = var_4813_begin_0, end = var_4813_end_0, end_mask = var_4813_end_mask_0, x = q_65_cast_fp16)[name = string("op_4813_cast_fp16")]; + bool q_interleave_0 = const()[name = string("q_interleave_0"), val = bool(false)]; + tensor q_cast_fp16 = concat(axis = var_4727, interleave = q_interleave_0, values = (var_4812_cast_fp16, var_4813_cast_fp16))[name = string("q_cast_fp16")]; + tensor var_4816_begin_0 = const()[name = string("op_4816_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4816_end_0 = const()[name = string("op_4816_end_0"), val = tensor([1, 4, 1024, 64])]; + tensor var_4816_end_mask_0 = const()[name = string("op_4816_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_55_cast_fp16 = transpose(perm = k_55_perm_0, x = var_4787_cast_fp16)[name = string("transpose_2")]; + tensor var_4816_cast_fp16 = slice_by_index(begin = var_4816_begin_0, end = var_4816_end_0, end_mask = var_4816_end_mask_0, x = k_55_cast_fp16)[name = string("op_4816_cast_fp16")]; + tensor var_4817_cast_fp16 = mul(x = var_4816_cast_fp16, y = c_1_cast_fp16)[name = string("op_4817_cast_fp16")]; + tensor var_4824_begin_0 = const()[name = string("op_4824_begin_0"), val = tensor([0, 0, 0, 32])]; + tensor var_4824_end_0 = const()[name = string("op_4824_end_0"), val = tensor([1, 4, 1024, 64])]; + tensor var_4824_end_mask_0 = const()[name = string("op_4824_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4824_cast_fp16 = slice_by_index(begin = var_4824_begin_0, end = var_4824_end_0, end_mask = var_4824_end_mask_0, x = var_4816_cast_fp16)[name = string("op_4824_cast_fp16")]; + fp16 const_81_promoted_to_fp16 = const()[name = string("const_81_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_4825_cast_fp16 = mul(x = var_4824_cast_fp16, y = const_81_promoted_to_fp16)[name = string("op_4825_cast_fp16")]; + tensor var_4826_begin_0 = const()[name = string("op_4826_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4826_end_0 = const()[name = string("op_4826_end_0"), val = tensor([1, 4, 1024, 32])]; + tensor var_4826_end_mask_0 = const()[name = string("op_4826_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4826_cast_fp16 = slice_by_index(begin = var_4826_begin_0, end = var_4826_end_0, end_mask = var_4826_end_mask_0, x = var_4816_cast_fp16)[name = string("op_4826_cast_fp16")]; + bool var_4828_interleave_0 = const()[name = string("op_4828_interleave_0"), val = bool(false)]; + tensor var_4828_cast_fp16 = concat(axis = var_4727, interleave = var_4828_interleave_0, values = (var_4825_cast_fp16, var_4826_cast_fp16))[name = string("op_4828_cast_fp16")]; + tensor var_4829_cast_fp16 = mul(x = var_4828_cast_fp16, y = s_1_cast_fp16)[name = string("op_4829_cast_fp16")]; + tensor var_4830_cast_fp16 = add(x = var_4817_cast_fp16, y = var_4829_cast_fp16)[name = string("op_4830_cast_fp16")]; + tensor var_4831_begin_0 = const()[name = string("op_4831_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_4831_end_0 = const()[name = string("op_4831_end_0"), val = tensor([1, 4, 1024, 256])]; + tensor var_4831_end_mask_0 = const()[name = string("op_4831_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4831_cast_fp16 = slice_by_index(begin = var_4831_begin_0, end = var_4831_end_0, end_mask = var_4831_end_mask_0, x = k_55_cast_fp16)[name = string("op_4831_cast_fp16")]; + bool k_57_interleave_0 = const()[name = string("k_57_interleave_0"), val = bool(false)]; + tensor k_57_cast_fp16 = concat(axis = var_4727, interleave = k_57_interleave_0, values = (var_4830_cast_fp16, var_4831_cast_fp16))[name = string("k_57_cast_fp16")]; + tensor var_4836_axes_0 = const()[name = string("op_4836_axes_0"), val = tensor([2])]; + tensor var_4836_cast_fp16 = expand_dims(axes = var_4836_axes_0, x = k_57_cast_fp16)[name = string("op_4836_cast_fp16")]; + tensor var_4838_reps_0 = const()[name = string("op_4838_reps_0"), val = tensor([1, 1, 4, 1, 1])]; + tensor var_4838_cast_fp16 = tile(reps = var_4838_reps_0, x = var_4836_cast_fp16)[name = string("op_4838_cast_fp16")]; + tensor var_4839 = const()[name = string("op_4839"), val = tensor([1, 16, 1024, 256])]; + tensor k_cast_fp16 = reshape(shape = var_4839, x = var_4838_cast_fp16)[name = string("k_cast_fp16")]; + tensor var_4843_axes_0 = const()[name = string("op_4843_axes_0"), val = tensor([2])]; + tensor v_29_cast_fp16 = transpose(perm = v_29_perm_0, x = var_4792_cast_fp16)[name = string("transpose_1")]; + tensor var_4843_cast_fp16 = expand_dims(axes = var_4843_axes_0, x = v_29_cast_fp16)[name = string("op_4843_cast_fp16")]; + tensor var_4845_reps_0 = const()[name = string("op_4845_reps_0"), val = tensor([1, 1, 4, 1, 1])]; + tensor var_4845_cast_fp16 = tile(reps = var_4845_reps_0, x = var_4843_cast_fp16)[name = string("op_4845_cast_fp16")]; + tensor var_4846 = const()[name = string("op_4846"), val = tensor([1, 16, 1024, 256])]; + tensor v_cast_fp16 = reshape(shape = var_4846, x = var_4845_cast_fp16)[name = string("v_cast_fp16")]; + bool var_4849_transpose_x_1 = const()[name = string("op_4849_transpose_x_1"), val = bool(false)]; + bool var_4849_transpose_y_1 = const()[name = string("op_4849_transpose_y_1"), val = bool(true)]; + tensor var_4849_cast_fp16 = matmul(transpose_x = var_4849_transpose_x_1, transpose_y = var_4849_transpose_y_1, x = q_cast_fp16, y = k_cast_fp16)[name = string("op_4849_cast_fp16")]; + fp16 var_4850_to_fp16 = const()[name = string("op_4850_to_fp16"), val = fp16(0x1p-4)]; + tensor var_4851_cast_fp16 = mul(x = var_4849_cast_fp16, y = var_4850_to_fp16)[name = string("op_4851_cast_fp16")]; + tensor scores_cast_fp16 = add(x = var_4851_cast_fp16, y = mask_to_fp16)[name = string("scores_cast_fp16")]; + tensor var_4853_cast_fp16 = softmax(axis = var_4727, x = scores_cast_fp16)[name = string("op_4853_cast_fp16")]; + bool out_5_transpose_x_0 = const()[name = string("out_5_transpose_x_0"), val = bool(false)]; + bool out_5_transpose_y_0 = const()[name = string("out_5_transpose_y_0"), val = bool(false)]; + tensor out_5_cast_fp16 = matmul(transpose_x = out_5_transpose_x_0, transpose_y = out_5_transpose_y_0, x = var_4853_cast_fp16, y = v_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor var_4855_perm_0 = const()[name = string("op_4855_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor var_4856 = const()[name = string("op_4856"), val = tensor([1, 1024, 4096])]; + tensor var_4855_cast_fp16 = transpose(perm = var_4855_perm_0, x = out_5_cast_fp16)[name = string("transpose_0")]; + tensor out_cast_fp16 = reshape(shape = var_4856, x = var_4855_cast_fp16)[name = string("out_cast_fp16")]; + tensor var_4858_cast_fp16 = sigmoid(x = gate_cast_fp16)[name = string("op_4858_cast_fp16")]; + tensor input_105_cast_fp16 = mul(x = out_cast_fp16, y = var_4858_cast_fp16)[name = string("input_105_cast_fp16")]; + tensor layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1628386368)))]; + tensor linear_58_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_58_cast_fp16")]; + tensor x_195_cast_fp16 = add(x = x_177_cast_fp16, y = linear_58_cast_fp16)[name = string("x_195_cast_fp16")]; + fp16 var_4726_promoted_3_to_fp16 = const()[name = string("op_4726_promoted_3_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4864_cast_fp16 = pow(x = x_195_cast_fp16, y = var_4726_promoted_3_to_fp16)[name = string("op_4864_cast_fp16")]; + tensor var_4866_axes_0 = const()[name = string("op_4866_axes_0"), val = tensor([-1])]; + bool var_4866_keep_dims_0 = const()[name = string("op_4866_keep_dims_0"), val = bool(true)]; + tensor var_4866_cast_fp16 = reduce_mean(axes = var_4866_axes_0, keep_dims = var_4866_keep_dims_0, x = var_4864_cast_fp16)[name = string("op_4866_cast_fp16")]; + fp16 var_4867_to_fp16 = const()[name = string("op_4867_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4868_cast_fp16 = add(x = var_4866_cast_fp16, y = var_4867_to_fp16)[name = string("op_4868_cast_fp16")]; + fp32 var_4869_epsilon_0 = const()[name = string("op_4869_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4869_cast_fp16 = rsqrt(epsilon = var_4869_epsilon_0, x = var_4868_cast_fp16)[name = string("op_4869_cast_fp16")]; + tensor x_197_cast_fp16 = mul(x = x_195_cast_fp16, y = var_4869_cast_fp16)[name = string("x_197_cast_fp16")]; + tensor var_4872_to_fp16 = const()[name = string("op_4872_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1649357952)))]; + tensor input_107_cast_fp16 = mul(x = x_197_cast_fp16, y = var_4872_to_fp16)[name = string("input_107_cast_fp16")]; + tensor layers_7_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1649363136)))]; + tensor linear_59_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_7_mlp_gate_proj_weight_to_fp16, x = input_107_cast_fp16)[name = string("linear_59_cast_fp16")]; + tensor var_4879_cast_fp16 = silu(x = linear_59_cast_fp16)[name = string("op_4879_cast_fp16")]; + tensor layers_7_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1696549120)))]; + tensor linear_60_cast_fp16 = linear(bias = linear_5_bias_0_to_fp16, weight = layers_7_mlp_up_proj_weight_to_fp16, x = input_107_cast_fp16)[name = string("linear_60_cast_fp16")]; + tensor input_111_cast_fp16 = mul(x = var_4879_cast_fp16, y = linear_60_cast_fp16)[name = string("input_111_cast_fp16")]; + tensor layers_7_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1743735104)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_7_mlp_down_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_61_cast_fp16")]; + tensor hidden_cast_fp16 = add(x = x_195_cast_fp16, y = linear_61_cast_fp16)[name = string("hidden_cast_fp16")]; + tensor var_4888_begin_0 = const()[name = string("op_4888_begin_0"), val = tensor([0, 0, 0])]; + tensor var_4888_end_0 = const()[name = string("op_4888_end_0"), val = tensor([1, 1024, 2560])]; + tensor var_4888_end_mask_0 = const()[name = string("op_4888_end_mask_0"), val = tensor([false, true, true])]; + tensor var_4888_squeeze_mask_0 = const()[name = string("op_4888_squeeze_mask_0"), val = tensor([true, false, false])]; + tensor var_4888_cast_fp16 = slice_by_index(begin = var_4888_begin_0, end = var_4888_end_0, end_mask = var_4888_end_mask_0, squeeze_mask = var_4888_squeeze_mask_0, x = hidden_cast_fp16)[name = string("op_4888_cast_fp16")]; + bool x_199_transpose_x_0 = const()[name = string("x_199_transpose_x_0"), val = bool(false)]; + bool x_199_transpose_y_0 = const()[name = string("x_199_transpose_y_0"), val = bool(false)]; + tensor x_199_cast_fp16 = matmul(transpose_x = x_199_transpose_x_0, transpose_y = x_199_transpose_y_0, x = last_onehot, y = var_4888_cast_fp16)[name = string("x_199_cast_fp16")]; + fp16 var_4894_promoted_to_fp16 = const()[name = string("op_4894_promoted_to_fp16"), val = fp16(0x1p+1)]; + tensor var_4896_cast_fp16 = pow(x = x_199_cast_fp16, y = var_4894_promoted_to_fp16)[name = string("op_4896_cast_fp16")]; + tensor var_4898_axes_0 = const()[name = string("op_4898_axes_0"), val = tensor([-1])]; + bool var_4898_keep_dims_0 = const()[name = string("op_4898_keep_dims_0"), val = bool(true)]; + tensor var_4898_cast_fp16 = reduce_mean(axes = var_4898_axes_0, keep_dims = var_4898_keep_dims_0, x = var_4896_cast_fp16)[name = string("op_4898_cast_fp16")]; + fp16 var_4899_to_fp16 = const()[name = string("op_4899_to_fp16"), val = fp16(0x1.1p-20)]; + tensor var_4900_cast_fp16 = add(x = var_4898_cast_fp16, y = var_4899_to_fp16)[name = string("op_4900_cast_fp16")]; + fp32 var_4901_epsilon_0 = const()[name = string("op_4901_epsilon_0"), val = fp32(0x1.197998p-40)]; + tensor var_4901_cast_fp16 = rsqrt(epsilon = var_4901_epsilon_0, x = var_4900_cast_fp16)[name = string("op_4901_cast_fp16")]; + tensor x_cast_fp16 = mul(x = x_199_cast_fp16, y = var_4901_cast_fp16)[name = string("x_cast_fp16")]; + tensor var_4904_to_fp16 = const()[name = string("op_4904_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1790921088)))]; + tensor input_cast_fp16 = mul(x = x_cast_fp16, y = var_4904_to_fp16)[name = string("input_cast_fp16")]; + tensor letter_head_weight_to_fp16 = const()[name = string("letter_head_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1790926272)))]; + tensor linear_62_bias_0_to_fp16 = const()[name = string("linear_62_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1791059456)))]; + tensor letter_logits = linear(bias = linear_62_bias_0_to_fp16, weight = letter_head_weight_to_fp16, x = input_cast_fp16)[name = string("linear_62_cast_fp16")]; + } -> (letter_logits); +} \ No newline at end of file