{ "quantization": { "method": "jang-importance", "profile": "JANG_2L", "target_bits": 2.0, "actual_bits": 3.73, "block_size": 64, "calibration_method": "weights", "quantization_method": "mse", "scoring_method": "weight-magnitude", "bit_widths_used": [ 2, 3, 4, 6, 8 ], "passthrough_bit_widths_used": [ 16 ], "passthrough_tensor_count": 115, "quantization_scheme": "asymmetric", "quantization_backend": "mx.quantize", "hadamard_rotation": false, "mlp_asymmetry_floor": true, "split_gate_up_quant": false, "split_gate_bits": null, "split_up_bits": null, "n2_down_bits": null, "n2_linear_attn_input_bits": null, "n2_linear_attn_out_bits": null, "n2_self_attn_bits": null, "n2_shared_expert_gate_up_bits": null, "n2_token_io_bits": null, "n2_lm_head_bits": null, "expert_prune_keep": null, "expert_prune_map": null, "expert_prune_method": null, "tensor_quantization_manifest_schema": 1, "tensor_quantization_manifest_count": 173, "tensor_quantization_manifest": { "backbone.embeddings": { "source_tensor": "backbone.embeddings.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.embeddings.weight", "scales_key": "backbone.embeddings.scales", "biases_key": "backbone.embeddings.biases", "weight_shape": [ 131072, 504 ], "scales_shape": [ 131072, 42 ], "biases_shape": [ 131072, 42 ] }, "backbone.layers.0.mixer.in_proj": { "source_tensor": "backbone.layers.0.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.0.mixer.in_proj.weight", "scales_key": "backbone.layers.0.mixer.in_proj.scales", "biases_key": "backbone.layers.0.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.0.mixer.out_proj": { "source_tensor": "backbone.layers.0.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.0.mixer.out_proj.weight", "scales_key": "backbone.layers.0.mixer.out_proj.scales", "biases_key": "backbone.layers.0.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.1.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.1.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.1.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.1.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.1.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.1.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.1.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.1.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.1.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.1.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.1.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.1.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.1.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.1.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.1.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.1.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.1.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.1.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.1.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.1.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.2.mixer.in_proj": { "source_tensor": "backbone.layers.2.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.2.mixer.in_proj.weight", "scales_key": "backbone.layers.2.mixer.in_proj.scales", "biases_key": "backbone.layers.2.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.2.mixer.out_proj": { "source_tensor": "backbone.layers.2.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.2.mixer.out_proj.weight", "scales_key": "backbone.layers.2.mixer.out_proj.scales", "biases_key": "backbone.layers.2.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.3.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.3.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.3.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.3.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.3.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.3.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.3.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.3.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.3.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.3.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.3.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.3.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.3.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.3.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.3.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.3.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.3.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.3.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.3.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.3.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.4.mixer.in_proj": { "source_tensor": "backbone.layers.4.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.4.mixer.in_proj.weight", "scales_key": "backbone.layers.4.mixer.in_proj.scales", "biases_key": "backbone.layers.4.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.4.mixer.out_proj": { "source_tensor": "backbone.layers.4.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.4.mixer.out_proj.weight", "scales_key": "backbone.layers.4.mixer.out_proj.scales", "biases_key": "backbone.layers.4.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.5.mixer.k_proj": { "source_tensor": "backbone.layers.5.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.5.mixer.k_proj.weight", "scales_key": "backbone.layers.5.mixer.k_proj.scales", "biases_key": "backbone.layers.5.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.5.mixer.o_proj": { "source_tensor": "backbone.layers.5.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.5.mixer.o_proj.weight", "scales_key": "backbone.layers.5.mixer.o_proj.scales", "biases_key": "backbone.layers.5.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.5.mixer.q_proj": { "source_tensor": "backbone.layers.5.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.5.mixer.q_proj.weight", "scales_key": "backbone.layers.5.mixer.q_proj.scales", "biases_key": "backbone.layers.5.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.5.mixer.v_proj": { "source_tensor": "backbone.layers.5.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.5.mixer.v_proj.weight", "scales_key": "backbone.layers.5.mixer.v_proj.scales", "biases_key": "backbone.layers.5.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.6.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.6.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.6.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.6.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.6.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.6.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.6.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.6.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.6.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.6.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.6.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.6.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.6.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.6.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.6.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.6.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.6.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.6.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.6.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.6.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.7.mixer.in_proj": { "source_tensor": "backbone.layers.7.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.7.mixer.in_proj.weight", "scales_key": "backbone.layers.7.mixer.in_proj.scales", "biases_key": "backbone.layers.7.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.7.mixer.out_proj": { "source_tensor": "backbone.layers.7.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.7.mixer.out_proj.weight", "scales_key": "backbone.layers.7.mixer.out_proj.scales", "biases_key": "backbone.layers.7.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.10.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.10.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.10.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.10.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.10.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.10.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.10.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.10.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.10.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.10.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.10.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.10.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.10.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.10.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.10.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.10.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.10.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.10.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.10.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.10.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.11.mixer.in_proj": { "source_tensor": "backbone.layers.11.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.11.mixer.in_proj.weight", "scales_key": "backbone.layers.11.mixer.in_proj.scales", "biases_key": "backbone.layers.11.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.11.mixer.out_proj": { "source_tensor": "backbone.layers.11.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.11.mixer.out_proj.weight", "scales_key": "backbone.layers.11.mixer.out_proj.scales", "biases_key": "backbone.layers.11.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.12.mixer.k_proj": { "source_tensor": "backbone.layers.12.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.12.mixer.k_proj.weight", "scales_key": "backbone.layers.12.mixer.k_proj.scales", "biases_key": "backbone.layers.12.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.12.mixer.o_proj": { "source_tensor": "backbone.layers.12.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.12.mixer.o_proj.weight", "scales_key": "backbone.layers.12.mixer.o_proj.scales", "biases_key": "backbone.layers.12.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.12.mixer.q_proj": { "source_tensor": "backbone.layers.12.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.12.mixer.q_proj.weight", "scales_key": "backbone.layers.12.mixer.q_proj.scales", "biases_key": "backbone.layers.12.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.12.mixer.v_proj": { "source_tensor": "backbone.layers.12.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.12.mixer.v_proj.weight", "scales_key": "backbone.layers.12.mixer.v_proj.scales", "biases_key": "backbone.layers.12.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.8.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.8.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.8.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.8.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.8.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.8.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.8.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.8.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.8.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.8.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.8.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.8.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.8.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.8.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.8.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.8.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.8.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.8.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.8.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.8.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.9.mixer.in_proj": { "source_tensor": "backbone.layers.9.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.9.mixer.in_proj.weight", "scales_key": "backbone.layers.9.mixer.in_proj.scales", "biases_key": "backbone.layers.9.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.9.mixer.out_proj": { "source_tensor": "backbone.layers.9.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.9.mixer.out_proj.weight", "scales_key": "backbone.layers.9.mixer.out_proj.scales", "biases_key": "backbone.layers.9.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.13.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.13.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.13.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.13.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.13.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.13.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.13.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.13.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.13.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.13.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.13.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.13.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.13.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.13.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.13.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.13.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.13.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.13.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.13.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.13.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.14.mixer.in_proj": { "source_tensor": "backbone.layers.14.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.14.mixer.in_proj.weight", "scales_key": "backbone.layers.14.mixer.in_proj.scales", "biases_key": "backbone.layers.14.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.14.mixer.out_proj": { "source_tensor": "backbone.layers.14.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.14.mixer.out_proj.weight", "scales_key": "backbone.layers.14.mixer.out_proj.scales", "biases_key": "backbone.layers.14.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.15.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.15.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.15.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.15.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.15.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.15.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.15.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.15.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.15.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.15.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.15.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.15.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.15.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.15.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.15.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.15.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.15.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.15.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.15.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.15.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.16.mixer.in_proj": { "source_tensor": "backbone.layers.16.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.16.mixer.in_proj.weight", "scales_key": "backbone.layers.16.mixer.in_proj.scales", "biases_key": "backbone.layers.16.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.16.mixer.out_proj": { "source_tensor": "backbone.layers.16.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.16.mixer.out_proj.weight", "scales_key": "backbone.layers.16.mixer.out_proj.scales", "biases_key": "backbone.layers.16.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.17.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.17.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.17.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.17.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.17.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.17.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.17.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.17.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.17.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.17.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.17.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.17.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.17.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.17.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.17.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.17.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.17.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.17.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.17.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.17.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.18.mixer.in_proj": { "source_tensor": "backbone.layers.18.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.18.mixer.in_proj.weight", "scales_key": "backbone.layers.18.mixer.in_proj.scales", "biases_key": "backbone.layers.18.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.18.mixer.out_proj": { "source_tensor": "backbone.layers.18.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.18.mixer.out_proj.weight", "scales_key": "backbone.layers.18.mixer.out_proj.scales", "biases_key": "backbone.layers.18.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.19.mixer.k_proj": { "source_tensor": "backbone.layers.19.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.19.mixer.k_proj.weight", "scales_key": "backbone.layers.19.mixer.k_proj.scales", "biases_key": "backbone.layers.19.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.19.mixer.o_proj": { "source_tensor": "backbone.layers.19.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.19.mixer.o_proj.weight", "scales_key": "backbone.layers.19.mixer.o_proj.scales", "biases_key": "backbone.layers.19.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.19.mixer.q_proj": { "source_tensor": "backbone.layers.19.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.19.mixer.q_proj.weight", "scales_key": "backbone.layers.19.mixer.q_proj.scales", "biases_key": "backbone.layers.19.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.19.mixer.v_proj": { "source_tensor": "backbone.layers.19.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.19.mixer.v_proj.weight", "scales_key": "backbone.layers.19.mixer.v_proj.scales", "biases_key": "backbone.layers.19.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.20.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.20.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.20.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.20.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.20.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.20.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.20.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.20.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.20.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.20.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.20.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.20.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.20.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.20.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.20.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.20.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.20.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.20.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.20.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.20.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.21.mixer.in_proj": { "source_tensor": "backbone.layers.21.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.21.mixer.in_proj.weight", "scales_key": "backbone.layers.21.mixer.in_proj.scales", "biases_key": "backbone.layers.21.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.21.mixer.out_proj": { "source_tensor": "backbone.layers.21.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.21.mixer.out_proj.weight", "scales_key": "backbone.layers.21.mixer.out_proj.scales", "biases_key": "backbone.layers.21.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.22.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.22.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.22.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.22.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.22.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.22.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.22.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.22.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.22.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.22.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.22.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.22.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.22.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.22.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.22.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.22.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.22.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.22.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.22.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.22.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.23.mixer.in_proj": { "source_tensor": "backbone.layers.23.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.23.mixer.in_proj.weight", "scales_key": "backbone.layers.23.mixer.in_proj.scales", "biases_key": "backbone.layers.23.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.23.mixer.out_proj": { "source_tensor": "backbone.layers.23.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.23.mixer.out_proj.weight", "scales_key": "backbone.layers.23.mixer.out_proj.scales", "biases_key": "backbone.layers.23.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.24.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.24.mixer.experts.127.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.24.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.24.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.24.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.24.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.24.mixer.experts.127.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.24.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.24.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.24.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.24.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.24.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.24.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.24.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.24.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.24.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.24.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.24.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.24.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.24.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.25.mixer.in_proj": { "source_tensor": "backbone.layers.25.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.25.mixer.in_proj.weight", "scales_key": "backbone.layers.25.mixer.in_proj.scales", "biases_key": "backbone.layers.25.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.25.mixer.out_proj": { "source_tensor": "backbone.layers.25.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.25.mixer.out_proj.weight", "scales_key": "backbone.layers.25.mixer.out_proj.scales", "biases_key": "backbone.layers.25.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.26.mixer.k_proj": { "source_tensor": "backbone.layers.26.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.26.mixer.k_proj.weight", "scales_key": "backbone.layers.26.mixer.k_proj.scales", "biases_key": "backbone.layers.26.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.26.mixer.o_proj": { "source_tensor": "backbone.layers.26.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.26.mixer.o_proj.weight", "scales_key": "backbone.layers.26.mixer.o_proj.scales", "biases_key": "backbone.layers.26.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.26.mixer.q_proj": { "source_tensor": "backbone.layers.26.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.26.mixer.q_proj.weight", "scales_key": "backbone.layers.26.mixer.q_proj.scales", "biases_key": "backbone.layers.26.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.26.mixer.v_proj": { "source_tensor": "backbone.layers.26.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.26.mixer.v_proj.weight", "scales_key": "backbone.layers.26.mixer.v_proj.scales", "biases_key": "backbone.layers.26.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.27.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.27.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.27.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.27.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.27.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.27.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.27.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.27.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.27.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.27.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.27.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.27.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.27.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.27.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.27.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.27.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.27.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.27.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.27.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.27.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.28.mixer.in_proj": { "source_tensor": "backbone.layers.28.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.28.mixer.in_proj.weight", "scales_key": "backbone.layers.28.mixer.in_proj.scales", "biases_key": "backbone.layers.28.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.28.mixer.out_proj": { "source_tensor": "backbone.layers.28.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.28.mixer.out_proj.weight", "scales_key": "backbone.layers.28.mixer.out_proj.scales", "biases_key": "backbone.layers.28.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.29.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.29.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.29.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.29.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.29.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.29.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.29.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.29.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.29.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.29.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.29.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.29.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.29.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.29.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.29.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.29.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.29.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.29.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.29.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.29.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.30.mixer.in_proj": { "source_tensor": "backbone.layers.30.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.30.mixer.in_proj.weight", "scales_key": "backbone.layers.30.mixer.in_proj.scales", "biases_key": "backbone.layers.30.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.30.mixer.out_proj": { "source_tensor": "backbone.layers.30.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.30.mixer.out_proj.weight", "scales_key": "backbone.layers.30.mixer.out_proj.scales", "biases_key": "backbone.layers.30.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.31.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.31.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.31.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.31.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.31.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.31.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.31.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.31.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.31.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.31.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.31.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.31.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.31.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.31.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.31.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.31.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.31.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.31.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.31.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.31.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.32.mixer.in_proj": { "source_tensor": "backbone.layers.32.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.32.mixer.in_proj.weight", "scales_key": "backbone.layers.32.mixer.in_proj.scales", "biases_key": "backbone.layers.32.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.32.mixer.out_proj": { "source_tensor": "backbone.layers.32.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.32.mixer.out_proj.weight", "scales_key": "backbone.layers.32.mixer.out_proj.scales", "biases_key": "backbone.layers.32.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.33.mixer.k_proj": { "source_tensor": "backbone.layers.33.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.33.mixer.k_proj.weight", "scales_key": "backbone.layers.33.mixer.k_proj.scales", "biases_key": "backbone.layers.33.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.33.mixer.o_proj": { "source_tensor": "backbone.layers.33.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.33.mixer.o_proj.weight", "scales_key": "backbone.layers.33.mixer.o_proj.scales", "biases_key": "backbone.layers.33.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.33.mixer.q_proj": { "source_tensor": "backbone.layers.33.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.33.mixer.q_proj.weight", "scales_key": "backbone.layers.33.mixer.q_proj.scales", "biases_key": "backbone.layers.33.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.33.mixer.v_proj": { "source_tensor": "backbone.layers.33.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.33.mixer.v_proj.weight", "scales_key": "backbone.layers.33.mixer.v_proj.scales", "biases_key": "backbone.layers.33.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.34.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.34.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.34.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.34.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.34.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.34.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.34.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.34.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.34.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.34.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.34.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.34.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.34.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.34.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.34.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.34.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.34.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.34.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.34.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.34.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.35.mixer.in_proj": { "source_tensor": "backbone.layers.35.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.35.mixer.in_proj.weight", "scales_key": "backbone.layers.35.mixer.in_proj.scales", "biases_key": "backbone.layers.35.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.35.mixer.out_proj": { "source_tensor": "backbone.layers.35.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.35.mixer.out_proj.weight", "scales_key": "backbone.layers.35.mixer.out_proj.scales", "biases_key": "backbone.layers.35.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.36.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.36.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.36.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.36.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.36.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.36.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.36.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.36.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.36.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.36.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.36.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.36.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.36.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.36.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.36.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.36.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.36.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.36.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.36.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.36.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.37.mixer.in_proj": { "source_tensor": "backbone.layers.37.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.37.mixer.in_proj.weight", "scales_key": "backbone.layers.37.mixer.in_proj.scales", "biases_key": "backbone.layers.37.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.37.mixer.out_proj": { "source_tensor": "backbone.layers.37.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.37.mixer.out_proj.weight", "scales_key": "backbone.layers.37.mixer.out_proj.scales", "biases_key": "backbone.layers.37.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.38.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.38.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.38.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.38.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.38.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.38.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.38.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.38.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.38.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.38.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.38.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.38.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.38.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.38.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.38.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.38.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.38.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.38.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.38.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.38.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.39.mixer.in_proj": { "source_tensor": "backbone.layers.39.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.39.mixer.in_proj.weight", "scales_key": "backbone.layers.39.mixer.in_proj.scales", "biases_key": "backbone.layers.39.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.39.mixer.out_proj": { "source_tensor": "backbone.layers.39.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.39.mixer.out_proj.weight", "scales_key": "backbone.layers.39.mixer.out_proj.scales", "biases_key": "backbone.layers.39.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.40.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.40.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.40.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.40.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.40.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.40.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.40.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.40.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.40.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.40.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.40.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.40.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.40.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.40.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.40.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.40.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.40.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.40.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.40.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.40.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.41.mixer.in_proj": { "source_tensor": "backbone.layers.41.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.41.mixer.in_proj.weight", "scales_key": "backbone.layers.41.mixer.in_proj.scales", "biases_key": "backbone.layers.41.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.41.mixer.out_proj": { "source_tensor": "backbone.layers.41.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.41.mixer.out_proj.weight", "scales_key": "backbone.layers.41.mixer.out_proj.scales", "biases_key": "backbone.layers.41.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.42.mixer.k_proj": { "source_tensor": "backbone.layers.42.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.42.mixer.k_proj.weight", "scales_key": "backbone.layers.42.mixer.k_proj.scales", "biases_key": "backbone.layers.42.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.42.mixer.o_proj": { "source_tensor": "backbone.layers.42.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.42.mixer.o_proj.weight", "scales_key": "backbone.layers.42.mixer.o_proj.scales", "biases_key": "backbone.layers.42.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.42.mixer.q_proj": { "source_tensor": "backbone.layers.42.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.42.mixer.q_proj.weight", "scales_key": "backbone.layers.42.mixer.q_proj.scales", "biases_key": "backbone.layers.42.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "backbone.layers.42.mixer.v_proj": { "source_tensor": "backbone.layers.42.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.42.mixer.v_proj.weight", "scales_key": "backbone.layers.42.mixer.v_proj.scales", "biases_key": "backbone.layers.42.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "backbone.layers.43.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.43.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.43.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.43.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.43.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.43.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.43.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.43.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.43.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.43.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.43.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.43.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.43.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.43.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.43.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.43.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.43.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.43.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.43.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.43.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.44.mixer.in_proj": { "source_tensor": "backbone.layers.44.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.44.mixer.in_proj.weight", "scales_key": "backbone.layers.44.mixer.in_proj.scales", "biases_key": "backbone.layers.44.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.44.mixer.out_proj": { "source_tensor": "backbone.layers.44.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.44.mixer.out_proj.weight", "scales_key": "backbone.layers.44.mixer.out_proj.scales", "biases_key": "backbone.layers.44.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.45.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.45.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.45.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.45.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.45.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.45.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.45.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.45.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.45.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.45.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.45.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.45.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.45.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.45.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.45.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.45.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.45.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.45.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.45.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.45.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.46.mixer.in_proj": { "source_tensor": "backbone.layers.46.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.46.mixer.in_proj.weight", "scales_key": "backbone.layers.46.mixer.in_proj.scales", "biases_key": "backbone.layers.46.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.46.mixer.out_proj": { "source_tensor": "backbone.layers.46.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.46.mixer.out_proj.weight", "scales_key": "backbone.layers.46.mixer.out_proj.scales", "biases_key": "backbone.layers.46.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.47.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.47.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.47.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.47.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.47.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.47.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.47.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.47.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.47.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.47.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.47.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.47.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.47.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.47.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.47.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.47.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.47.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.47.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.47.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.47.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.48.mixer.in_proj": { "source_tensor": "backbone.layers.48.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.48.mixer.in_proj.weight", "scales_key": "backbone.layers.48.mixer.in_proj.scales", "biases_key": "backbone.layers.48.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.48.mixer.out_proj": { "source_tensor": "backbone.layers.48.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.48.mixer.out_proj.weight", "scales_key": "backbone.layers.48.mixer.out_proj.scales", "biases_key": "backbone.layers.48.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.49.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.49.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.49.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.49.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.49.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.49.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.49.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.49.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.49.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.49.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.49.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.49.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.49.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.49.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.49.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.49.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.49.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.49.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.49.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.49.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "backbone.layers.50.mixer.in_proj": { "source_tensor": "backbone.layers.50.mixer.in_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.50.mixer.in_proj.weight", "scales_key": "backbone.layers.50.mixer.in_proj.scales", "biases_key": "backbone.layers.50.mixer.in_proj.biases", "weight_shape": [ 10304, 504 ], "scales_shape": [ 10304, 42 ], "biases_shape": [ 10304, 42 ] }, "backbone.layers.50.mixer.out_proj": { "source_tensor": "backbone.layers.50.mixer.out_proj.weight", "bits": 6, "group_size": 64, "weight_key": "backbone.layers.50.mixer.out_proj.weight", "scales_key": "backbone.layers.50.mixer.out_proj.scales", "biases_key": "backbone.layers.50.mixer.out_proj.biases", "weight_shape": [ 2688, 768 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "backbone.layers.51.mixer.switch_mlp.fc2": { "source_tensor": "backbone.layers.51.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "backbone.layers.51.mixer.switch_mlp.fc2.weight", "scales_key": "backbone.layers.51.mixer.switch_mlp.fc2.scales", "biases_key": "backbone.layers.51.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "backbone.layers.51.mixer.switch_mlp.fc1": { "source_tensor": "backbone.layers.51.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "backbone.layers.51.mixer.switch_mlp.fc1.weight", "scales_key": "backbone.layers.51.mixer.switch_mlp.fc1.scales", "biases_key": "backbone.layers.51.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "backbone.layers.51.mixer.shared_experts.down_proj": { "source_tensor": "backbone.layers.51.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.51.mixer.shared_experts.down_proj.weight", "scales_key": "backbone.layers.51.mixer.shared_experts.down_proj.scales", "biases_key": "backbone.layers.51.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "backbone.layers.51.mixer.shared_experts.up_proj": { "source_tensor": "backbone.layers.51.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "backbone.layers.51.mixer.shared_experts.up_proj.weight", "scales_key": "backbone.layers.51.mixer.shared_experts.up_proj.scales", "biases_key": "backbone.layers.51.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] }, "lm_head": { "source_tensor": "lm_head.weight", "bits": 8, "group_size": 64, "weight_key": "lm_head.weight", "scales_key": "lm_head.scales", "biases_key": "lm_head.biases", "weight_shape": [ 131072, 672 ], "scales_shape": [ 131072, 42 ], "biases_shape": [ 131072, 42 ] }, "mtp.layers.0.eh_proj": { "source_tensor": "mtp.layers.0.eh_proj.weight", "bits": 2, "group_size": 64, "weight_key": "mtp.layers.0.eh_proj.weight", "scales_key": "mtp.layers.0.eh_proj.scales", "biases_key": "mtp.layers.0.eh_proj.biases", "weight_shape": [ 2688, 336 ], "scales_shape": [ 2688, 84 ], "biases_shape": [ 2688, 84 ] }, "mtp.layers.0.mixer.k_proj": { "source_tensor": "mtp.layers.0.mixer.k_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.0.mixer.k_proj.weight", "scales_key": "mtp.layers.0.mixer.k_proj.scales", "biases_key": "mtp.layers.0.mixer.k_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "mtp.layers.0.mixer.o_proj": { "source_tensor": "mtp.layers.0.mixer.o_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.0.mixer.o_proj.weight", "scales_key": "mtp.layers.0.mixer.o_proj.scales", "biases_key": "mtp.layers.0.mixer.o_proj.biases", "weight_shape": [ 2688, 1024 ], "scales_shape": [ 2688, 64 ], "biases_shape": [ 2688, 64 ] }, "mtp.layers.0.mixer.q_proj": { "source_tensor": "mtp.layers.0.mixer.q_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.0.mixer.q_proj.weight", "scales_key": "mtp.layers.0.mixer.q_proj.scales", "biases_key": "mtp.layers.0.mixer.q_proj.biases", "weight_shape": [ 4096, 672 ], "scales_shape": [ 4096, 42 ], "biases_shape": [ 4096, 42 ] }, "mtp.layers.0.mixer.v_proj": { "source_tensor": "mtp.layers.0.mixer.v_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.0.mixer.v_proj.weight", "scales_key": "mtp.layers.0.mixer.v_proj.scales", "biases_key": "mtp.layers.0.mixer.v_proj.biases", "weight_shape": [ 256, 672 ], "scales_shape": [ 256, 42 ], "biases_shape": [ 256, 42 ] }, "mtp.layers.1.mixer.switch_mlp.fc2": { "source_tensor": "mtp.layers.1.mixer.experts.99.down_proj.weight", "bits": 3, "group_size": 64, "weight_key": "mtp.layers.1.mixer.switch_mlp.fc2.weight", "scales_key": "mtp.layers.1.mixer.switch_mlp.fc2.scales", "biases_key": "mtp.layers.1.mixer.switch_mlp.fc2.biases", "weight_shape": [ 128, 2688, 174 ], "scales_shape": [ 128, 2688, 29 ], "biases_shape": [ 128, 2688, 29 ] }, "mtp.layers.1.mixer.switch_mlp.fc1": { "source_tensor": "mtp.layers.1.mixer.experts.99.up_proj.weight", "bits": 4, "group_size": 64, "weight_key": "mtp.layers.1.mixer.switch_mlp.fc1.weight", "scales_key": "mtp.layers.1.mixer.switch_mlp.fc1.scales", "biases_key": "mtp.layers.1.mixer.switch_mlp.fc1.biases", "weight_shape": [ 128, 1856, 336 ], "scales_shape": [ 128, 1856, 42 ], "biases_shape": [ 128, 1856, 42 ] }, "mtp.layers.1.mixer.shared_experts.down_proj": { "source_tensor": "mtp.layers.1.mixer.shared_experts.down_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.1.mixer.shared_experts.down_proj.weight", "scales_key": "mtp.layers.1.mixer.shared_experts.down_proj.scales", "biases_key": "mtp.layers.1.mixer.shared_experts.down_proj.biases", "weight_shape": [ 2688, 928 ], "scales_shape": [ 2688, 58 ], "biases_shape": [ 2688, 58 ] }, "mtp.layers.1.mixer.shared_experts.up_proj": { "source_tensor": "mtp.layers.1.mixer.shared_experts.up_proj.weight", "bits": 8, "group_size": 64, "weight_key": "mtp.layers.1.mixer.shared_experts.up_proj.weight", "scales_key": "mtp.layers.1.mixer.shared_experts.up_proj.scales", "biases_key": "mtp.layers.1.mixer.shared_experts.up_proj.biases", "weight_shape": [ 3712, 672 ], "scales_shape": [ 3712, 42 ], "biases_shape": [ 3712, 42 ] } } }, "source_model": { "name": "NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "dtype": "bfloat16", "parameters": "2.6B" }, "architecture": { "type": "moe", "attention": "gqa", "has_vision": false, "has_ssm": false, "has_moe": true }, "runtime": { "total_weight_bytes": 17415865096, "total_weight_gb": 16.22, "bundle_has_mtp": true, "mtp_layers": 1, "mtp_mode": "preserved_enabled" }, "mtp": { "artifact_available": true, "runtime_available": false, "num_nextn_predict_layers": 1, "shape": "deepseek_v3_eh_proj", "shares_embeddings_and_lm_head": true }, "bundle_has_mtp": true, "mtp_layers": 1, "capabilities": { "reasoning_parser": "deepseek_r1", "tool_parser": "nemotron", "think_in_template": true, "supports_tools": true, "supports_thinking": true, "family": "nemotron_h", "modality": "text", "modalities": { "text": true, "vision": false, "audio": false, "video": false }, "has_vision": false, "has_audio": false, "has_video": false, "cache_type": "hybrid", "default_reasoning": "on" }, "format": "jang", "format_version": "2.0", "chat": { "sampling_defaults": { "temperature": 1.0, "top_p": 0.95, "top_k": 0, "repetition_penalty": 1.0, "source": "vendor_card+generation_config_2026-08-11" }, "add_bos_token": false, "stop_token_ids": [ 2, 11 ] }, "reasoning": { "supported": true, "parser": "deepseek_r1", "default": "on", "think_in_template": true, "tiers": null, "budget": "server_side", "enable_kwarg": "enable_thinking", "off_is_prefilled_empty_block": true, "truncate_history_thinking_default": true }, "tools": { "supported": true, "parser": "nemotron", "dialect": "xml_function", "tools_in_system_prompt": true, "tool_results_role": "tool", "force_nonempty_content_supported": false } }