Add files using upload-large-folder tool

recipe/logs/C1_convert.log

@@ -0,0 +1,934 @@
+INFO:hf-to-gguf:Loading model: hf
+INFO:hf-to-gguf:Model architecture: Qwen3_5MoeForConditionalGeneration
+INFO:hf-to-gguf:gguf: loading model weight map from 'model.safetensors.index.json'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00001-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00002-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00003-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00004-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00005-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00006-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00007-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00008-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00009-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00010-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00011-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00012-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00013-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00014-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00015-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00016-of-00016.safetensors'
+INFO:gguf.gguf_writer:gguf: This GGUF file is for Little Endian only
+INFO:hf-to-gguf:Exporting model...
+INFO:hf-to-gguf:token_embd.weight,                    torch.bfloat16 --> BF16, shape = {2048, 248320}
+INFO:hf-to-gguf:blk.0.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.0.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.0.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.0.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.0.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.0.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.0.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.0.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.0.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.0.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.0.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.0.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.0.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.0.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.0.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.0.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.0.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.0.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.0.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.1.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.1.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.1.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.1.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.1.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.1.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.1.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.1.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.1.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.1.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.1.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.1.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.1.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.1.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.1.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.1.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.1.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.1.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.1.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.2.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.2.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.2.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.2.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.2.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.2.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.2.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.2.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.2.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.2.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.2.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.2.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.2.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.2.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.2.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.2.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.2.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.2.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.2.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.3.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.3.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.3.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.3.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.3.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.3.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.3.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.3.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.3.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.3.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.3.attn_k_norm.weight,             torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.attn_k.weight,                  torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.3.attn_output.weight,             torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.3.attn_q_norm.weight,             torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.attn_q.weight,                  torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.3.attn_v.weight,                  torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.4.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.4.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.4.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.4.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.4.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.4.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.4.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.4.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.4.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.4.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.4.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.4.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.4.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.4.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.4.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.4.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.4.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.4.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.4.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.5.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.5.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.5.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.5.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.5.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.5.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.5.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.5.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.5.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.5.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.5.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.5.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.5.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.5.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.5.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.5.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.5.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.5.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.5.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.6.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.6.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.6.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.6.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.6.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.6.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.6.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.6.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.6.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.6.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.6.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.6.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.6.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.6.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.6.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.6.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.6.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.6.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.6.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.7.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.7.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.7.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.7.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.7.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.7.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.7.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.7.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.7.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.7.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.7.attn_k_norm.weight,             torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.7.attn_k.weight,                  torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.7.attn_output.weight,             torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.7.attn_q_norm.weight,             torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.7.attn_q.weight,                  torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.7.attn_v.weight,                  torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.8.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.8.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.8.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.8.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.8.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.8.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.8.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.8.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.8.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.8.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.8.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.8.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.8.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.8.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.8.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.8.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.8.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.8.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.8.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.9.attn_norm.weight,               torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.9.ssm_a,                          torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.9.ssm_conv1d.weight,              torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.9.ssm_dt.bias,                    torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.9.ssm_alpha.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.9.ssm_beta.weight,                torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.9.attn_qkv.weight,                torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.9.attn_gate.weight,               torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.9.ssm_norm.weight,                torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.9.ssm_out.weight,                 torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.9.ffn_gate_exps.weight,           torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.9.ffn_up_exps.weight,             torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.9.ffn_gate_inp.weight,            torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.9.ffn_down_shexp.weight,          torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.9.ffn_gate_shexp.weight,          torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.9.ffn_up_shexp.weight,            torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.9.ffn_gate_inp_shexp.weight,      torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.10.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.10.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.10.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.10.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.10.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.10.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.10.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.10.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.10.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.10.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.10.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.10.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.10.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.10.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.10.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.10.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.10.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.10.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.10.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.11.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.11.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.11.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.11.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.11.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.11.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.11.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.11.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.11.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.11.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.11.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.11.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.11.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.11.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.11.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.11.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.12.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.12.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.12.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.12.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.12.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.12.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.12.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.12.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.12.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.12.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.12.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.12.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.12.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.12.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.12.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.9.ffn_down_exps.weight,           torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.9.post_attention_norm.weight,     torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.12.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.12.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.12.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.12.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.13.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.13.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.13.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.13.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.13.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.13.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.13.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.13.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.13.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.13.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.13.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.13.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.13.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.13.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.13.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.13.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.13.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.13.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.13.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.14.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.14.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.14.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.14.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.14.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.14.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.14.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.14.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.14.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.14.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.14.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.14.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.14.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.14.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.14.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.14.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.14.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.14.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.14.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.15.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.15.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.15.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.15.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.15.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.15.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.15.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.15.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.15.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.15.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.15.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.15.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.15.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.15.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.15.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.15.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.16.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.16.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.16.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.16.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.16.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.16.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.16.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.16.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.16.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.16.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.16.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.16.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.16.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.16.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.16.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.16.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.16.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.16.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.16.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.17.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.17.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.17.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.17.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.17.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.17.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.17.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.17.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.17.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.17.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.17.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.17.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.17.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.17.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.17.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.17.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.17.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.17.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.17.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.18.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.18.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.18.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.18.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.18.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.18.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.18.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.18.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.18.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.18.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.18.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.18.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.18.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.18.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.18.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.18.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.18.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.18.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.18.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.19.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.19.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.19.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.19.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.19.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.19.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.19.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.19.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.19.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.19.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.19.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.19.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.19.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.19.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.19.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.19.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.20.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.20.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.20.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.20.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.20.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.20.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.20.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.20.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.20.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.20.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.20.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.20.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.20.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.20.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.20.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.20.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.20.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.20.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.20.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.21.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.21.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.21.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.21.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.21.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.21.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.21.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.21.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.21.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.21.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.21.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.21.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.21.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.21.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.21.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.21.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.21.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.21.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.21.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.22.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.22.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.22.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.22.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.22.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.22.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.22.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.22.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.22.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.22.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.22.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.22.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.22.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.22.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.22.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.22.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.22.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.22.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.22.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.23.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.23.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.23.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.23.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.23.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.23.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.23.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.23.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.23.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.23.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.23.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.23.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.23.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.23.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.23.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.23.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.24.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.24.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.24.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.24.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.24.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.24.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.24.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.24.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.24.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.24.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.24.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.24.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.24.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.24.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.24.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.24.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.24.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.24.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.24.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.25.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.25.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.25.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.25.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.25.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.25.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.25.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.25.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.25.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.25.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.25.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.25.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.25.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.25.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.25.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.25.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.25.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.25.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.25.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.26.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.26.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.26.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.26.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.26.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.26.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.26.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.26.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.26.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.26.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.26.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.26.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.26.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.26.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.26.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.26.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.26.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.26.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.26.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.27.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.27.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.27.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.27.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.27.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.27.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.27.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.27.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.27.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.27.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.27.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.27.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.27.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.27.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.27.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.27.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.28.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.28.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.28.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.28.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.28.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.28.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.28.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.28.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.28.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.28.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.28.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.28.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.28.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.28.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.28.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.28.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.28.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.28.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.28.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.29.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.29.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.29.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.29.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.29.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.29.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.29.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.29.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.29.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.29.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.29.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.29.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.29.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.29.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.29.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.29.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.29.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.29.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.29.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.30.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.30.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.30.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.30.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.30.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.30.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.30.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.30.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.30.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.30.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.30.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.30.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.30.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.30.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.30.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.30.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.30.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.30.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.30.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.31.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.31.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.31.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.31.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.31.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.31.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.31.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.31.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.31.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.31.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.31.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.31.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.31.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.31.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.31.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.31.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.32.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.32.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.32.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.32.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.32.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.32.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.32.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.32.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.32.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.32.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.32.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.32.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.32.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.32.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.32.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.32.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.32.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.32.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.32.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.33.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.33.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.33.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.33.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.33.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.33.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.33.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.33.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.33.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.33.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.33.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.33.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.33.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.33.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.33.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.33.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.33.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.33.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.33.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.34.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.34.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.34.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.34.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.34.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.34.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.34.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.34.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.34.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.34.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.34.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.34.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.34.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.34.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.34.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.34.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.34.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.34.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.34.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.35.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.35.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.35.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.35.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.35.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.35.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.35.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.35.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.35.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.35.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.35.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.35.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.35.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.35.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.35.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.35.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.36.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.36.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.36.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.36.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.36.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.36.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.36.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.36.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.36.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.36.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.36.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.36.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.36.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.36.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.36.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.36.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.36.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.36.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.36.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.37.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.37.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.37.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.37.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.37.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.37.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.37.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.37.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.37.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.37.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.37.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.37.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.37.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.37.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.37.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.37.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.37.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.37.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.37.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.38.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.38.ssm_a,                         torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.38.ssm_conv1d.weight,             torch.bfloat16 --> F32, shape = {4, 8192}
+INFO:hf-to-gguf:blk.38.ssm_dt.bias,                   torch.bfloat16 --> F32, shape = {32}
+INFO:hf-to-gguf:blk.38.ssm_alpha.weight,              torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.38.ssm_beta.weight,               torch.bfloat16 --> BF16, shape = {2048, 32}
+INFO:hf-to-gguf:blk.38.attn_qkv.weight,               torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.38.attn_gate.weight,              torch.bfloat16 --> BF16, shape = {2048, 4096}
+INFO:hf-to-gguf:blk.38.ssm_norm.weight,               torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.38.ssm_out.weight,                torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.38.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.38.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.38.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.38.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.38.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.38.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.38.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.38.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.38.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:output.weight,                        torch.bfloat16 --> BF16, shape = {2048, 248320}
+INFO:hf-to-gguf:blk.39.attn_norm.weight,              torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.39.ffn_down_exps.weight,          torch.bfloat16 --> BF16, shape = {512, 2048, 256}
+INFO:hf-to-gguf:blk.39.ffn_gate_exps.weight,          torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.39.ffn_up_exps.weight,            torch.bfloat16 --> BF16, shape = {2048, 512, 256}
+INFO:hf-to-gguf:blk.39.ffn_gate_inp.weight,           torch.bfloat16 --> F32, shape = {2048, 256}
+INFO:hf-to-gguf:blk.39.ffn_down_shexp.weight,         torch.bfloat16 --> BF16, shape = {512, 2048}
+INFO:hf-to-gguf:blk.39.ffn_gate_shexp.weight,         torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.39.ffn_up_shexp.weight,           torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.39.ffn_gate_inp_shexp.weight,     torch.bfloat16 --> F32, shape = {2048, 1}
+INFO:hf-to-gguf:blk.39.post_attention_norm.weight,    torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:blk.39.attn_k_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.39.attn_k.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:blk.39.attn_output.weight,            torch.bfloat16 --> BF16, shape = {4096, 2048}
+INFO:hf-to-gguf:blk.39.attn_q_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.39.attn_q.weight,                 torch.bfloat16 --> BF16, shape = {2048, 8192}
+INFO:hf-to-gguf:blk.39.attn_v.weight,                 torch.bfloat16 --> BF16, shape = {2048, 512}
+INFO:hf-to-gguf:output_norm.weight,                   torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:Set meta model
+INFO:hf-to-gguf:Set model parameters
+INFO:hf-to-gguf:gguf: context length = 262144
+INFO:hf-to-gguf:gguf: embedding length = 2048
+INFO:hf-to-gguf:gguf: head count = 16
+INFO:hf-to-gguf:gguf: key-value head count = 2
+WARNING:hf-to-gguf:Unknown RoPE type: default
+INFO:hf-to-gguf:gguf: rope scaling type = NONE
+INFO:hf-to-gguf:gguf: mrope sections: [11, 11, 10, 0]
+INFO:hf-to-gguf:gguf: rope theta = 10000000
+INFO:hf-to-gguf:gguf: rms norm epsilon = 1e-06
+INFO:hf-to-gguf:gguf: expert count = 256
+INFO:hf-to-gguf:gguf: experts used count = 8
+INFO:hf-to-gguf:gguf: file type = 32
+INFO:hf-to-gguf:gguf: expert feed forward length = 512
+INFO:hf-to-gguf:gguf: expert shared feed forward length = 512
+INFO:hf-to-gguf:Set model quantization version
+INFO:hf-to-gguf:Set model tokenizer
+INFO:gguf.vocab:Adding 247587 merge(s).
+INFO:gguf.vocab:Setting special token type eos to 248046
+INFO:gguf.vocab:Setting special token type pad to 248044
+INFO:gguf.vocab:Setting chat_template to {%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if reasoning_effort is not defined or reasoning_effort is none %}
+        {{- '<think>' }}
+    {%- elif reasoning_effort == 'none' %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- elif reasoning_effort == 'high' %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>' }}
+    {%- endif %}
+{%- endif %}
+
+INFO:gguf.gguf_writer:Writing the following files:
+INFO:gguf.gguf_writer:gguf/Nex-N2.5-mini-BF16.gguf: n_tensors = 733, total_size = 69.4G
+
+INFO:hf-to-gguf:Model successfully exported to gguf/Nex-N2.5-mini-BF16.gguf

recipe/logs/C2_mmproj.log

@@ -0,0 +1,362 @@
+INFO:hf-to-gguf:Loading model: hf
+INFO:hf-to-gguf:Model architecture: Qwen3_5MoeForConditionalGeneration
+INFO:hf-to-gguf:gguf: loading model weight map from 'model.safetensors.index.json'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00001-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00002-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00003-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00004-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00005-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00006-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00007-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00008-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00009-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00010-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00011-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00012-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00013-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00014-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00015-of-00016.safetensors'
+INFO:hf-to-gguf:gguf: indexing model part 'model-00016-of-00016.safetensors'
+INFO:gguf.gguf_writer:gguf: This GGUF file is for Little Endian only
+INFO:hf-to-gguf:Exporting model...
+INFO:hf-to-gguf:v.blk.0.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.0.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.0.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.0.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.0.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.0.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.0.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.0.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.0.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.0.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.0.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.0.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.1.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.1.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.1.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.1.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.1.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.1.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.1.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.10.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.10.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.10.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.10.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.10.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.10.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.10.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.11.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.11.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.11.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.11.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.11.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.11.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.11.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.12.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.12.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.12.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.12.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.12.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.12.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.12.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.13.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.13.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.13.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.13.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.13.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.13.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.13.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.14.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.14.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.14.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.14.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.14.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.14.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.14.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.15.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.15.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.15.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.15.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.15.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.15.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.15.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.16.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.16.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.16.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.16.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.16.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.16.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.16.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.17.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.17.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.17.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.17.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.17.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.17.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.17.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.18.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.18.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.18.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.18.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.18.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.18.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.18.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.19.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.19.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.19.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.19.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.19.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.19.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.19.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.2.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.2.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.2.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.2.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.2.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.2.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.2.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.20.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.20.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.20.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.20.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.20.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.20.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.20.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.21.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.21.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.21.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.21.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.21.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.21.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.21.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.22.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.22.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.22.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.22.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.22.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.22.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.22.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.23.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.23.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.23.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.23.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.23.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.23.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.23.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.24.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.24.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.24.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.24.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.24.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.24.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.24.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.25.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.25.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.25.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.25.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.25.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.25.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.25.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.attn_out.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.attn_out.weight,             torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.26.attn_qkv.bias,               torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.26.attn_qkv.weight,             torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.26.ffn_up.bias,                 torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.26.ffn_up.weight,               torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.26.ffn_down.bias,               torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.ffn_down.weight,             torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.26.ln1.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.ln1.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.ln2.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.26.ln2.weight,                  torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.3.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.3.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.3.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.3.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.3.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.3.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.3.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.4.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.4.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.4.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.4.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.4.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.4.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.4.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.5.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.5.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.5.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.5.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.5.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.5.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.5.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.6.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.6.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.6.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.6.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.6.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.6.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.6.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.7.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.7.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.7.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.7.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.7.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.7.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.7.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.8.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.8.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.8.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.8.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.8.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.8.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.8.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.attn_out.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.attn_out.weight,              torch.bfloat16 --> BF16, shape = {1152, 1152}
+INFO:hf-to-gguf:v.blk.9.attn_qkv.bias,                torch.bfloat16 --> F32, shape = {3456}
+INFO:hf-to-gguf:v.blk.9.attn_qkv.weight,              torch.bfloat16 --> BF16, shape = {1152, 3456}
+INFO:hf-to-gguf:v.blk.9.ffn_up.bias,                  torch.bfloat16 --> F32, shape = {4304}
+INFO:hf-to-gguf:v.blk.9.ffn_up.weight,                torch.bfloat16 --> BF16, shape = {1152, 4304}
+INFO:hf-to-gguf:v.blk.9.ffn_down.bias,                torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.ffn_down.weight,              torch.bfloat16 --> BF16, shape = {4304, 1152}
+INFO:hf-to-gguf:v.blk.9.ln1.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.ln1.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.ln2.bias,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.blk.9.ln2.weight,                   torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:mm.0.bias,                            torch.bfloat16 --> F32, shape = {4608}
+INFO:hf-to-gguf:mm.0.weight,                          torch.bfloat16 --> BF16, shape = {4608, 4608}
+INFO:hf-to-gguf:mm.2.bias,                            torch.bfloat16 --> F32, shape = {2048}
+INFO:hf-to-gguf:mm.2.weight,                          torch.bfloat16 --> BF16, shape = {4608, 2048}
+INFO:hf-to-gguf:v.post_ln.bias,                       torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.post_ln.weight,                     torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.patch_embd.bias,                    torch.bfloat16 --> F32, shape = {1152}
+INFO:hf-to-gguf:v.patch_embd.weight,                  torch.bfloat16 --> F32, shape = {16, 16, 3, 1152}
+INFO:hf-to-gguf:v.patch_embd.weight.1,                torch.bfloat16 --> F32, shape = {16, 16, 3, 1152}
+INFO:hf-to-gguf:v.position_embd.weight,               torch.bfloat16 --> F32, shape = {1152, 2304}
+INFO:hf-to-gguf:Set meta model
+INFO:hf-to-gguf:Set model parameters
+INFO:hf-to-gguf:Set model quantization version
+INFO:gguf.gguf_writer:Writing the following files:
+INFO:gguf.gguf_writer:out/mmproj-Nex-N2.5-mini-BF16.gguf: n_tensors = 334, total_size = 902.8M
+
+INFO:hf-to-gguf:Model successfully exported to out/mmproj-Nex-N2.5-mini-BF16.gguf

recipe/logs/C_readback.log

@@ -0,0 +1,2 @@
+PASS Nex-N2.5-mini-BF16.gguf arch=qwen35moe ftype=32 tensors=733 nextn=0 output.weight=BF16 token_embd.weight=BF16
+PASS mmproj-Nex-N2.5-mini-BF16.gguf arch=clip ftype=32 tensors=334 nextn=0 output.weight=None token_embd.weight=None

recipe/logs/D2_verify_download.log

@@ -0,0 +1,2 @@
+files=28 sha-verified=19 size-only=9 bad=0
+RESULT: PASS

recipe/logs/N1_ppl_bf16.log

@@ -0,0 +1,15 @@
+0.00.044.003 I common_init_result: fitting params to device memory ...
+0.00.044.006 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.378.699 W llama_model_loader: direct I/O is enabled, disabling mmap
+1.17.123.477 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+1.17.377.358 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+1.18.283.894 I 
+1.18.284.021 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+1.18.284.129 I perplexity: saving all logits to kld/bf16.kld
+1.18.284.134 I perplexity: tokenizing the input ..
+1.18.617.280 I perplexity: tokenization took 333.135 ms
+1.18.617.370 I perplexity: calculating perplexity over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+1.23.386.958 I perplexity: 4.58 seconds per pass - ETA 3.05 minutes
+[1]139.1466,[2]166.0748,[3]174.5814,[4]170.5873,[5]171.7312,[6]127.5767,[7]99.6563,[8]94.8236,[9]101.8520,[10]104.1594,[11]108.4984,[12]113.4569,[13]109.8859,[14]107.0003,[15]104.7124,[16]107.6891,[17]105.9352,[18]106.7826,[19]104.4693,[20]101.7712,[21]102.7759,[22]104.9236,[23]107.8566,[24]107.9859,[25]109.0207,[26]109.5885,[27]111.5203,[28]113.7558,[29]116.8379,[30]117.2211,[31]115.3159,[32]113.7423,[33]112.7430,[34]112.9143,[35]113.3873,[36]113.5581,[37]111.6761,[38]111.1729,[39]108.9264,[40]105.9103,
+4.26.432.497 I Final estimate: PPL = 105.9103 +/- 1.77389
+

recipe/logs/N1c_ppl_bf16_cpu.log

@@ -0,0 +1,15 @@
+0.00.040.030 I common_init_result: fitting params to device memory ...
+0.00.040.033 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.314.545 I common_params_fit_impl: projected to use 66756 MiB of host memory vs. 127438 MiB of total host memory
+0.00.652.877 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.00.672.925 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.01.325.340 I 
+0.01.325.484 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.01.326.996 I perplexity: saving all logits to kld/bf16.kld
+0.01.327.002 I perplexity: tokenizing the input ..
+0.01.631.676 I perplexity: tokenization took 304.664 ms
+0.01.631.766 I perplexity: calculating perplexity over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.13.030.135 I perplexity: 11.26 seconds per pass - ETA 7.50 minutes
+[1]5.6964,[2]6.6666,[3]7.0328,[4]7.2903,[5]7.1445,[6]6.1493,[7]5.7412,[8]5.6706,[9]5.9764,[10]6.0886,[11]6.1422,[12]6.3827,[13]6.4266,[14]6.4831,[15]6.5233,[16]6.6999,[17]6.7466,[18]6.8361,[19]6.7889,[20]6.5275,[21]6.5437,[22]6.5571,[23]6.6058,[24]6.6060,[25]6.6371,[26]6.6074,[27]6.7700,[28]6.8576,[29]6.8518,[30]6.7971,[31]6.6991,[32]6.6042,[33]6.5393,[34]6.5251,[35]6.5369,[36]6.5530,[37]6.4632,[38]6.3955,[39]6.3171,[40]6.2303,
+7.32.185.106 I Final estimate: PPL = 6.2303 +/- 0.07538
+

recipe/logs/N2c_imatrix_cpu.log

@@ -0,0 +1,114 @@
+0.00.047.060 I common_init_result: fitting params to device memory ...
+0.00.047.066 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.538.240 I common_params_fit_impl: projected to use 66727 MiB of host memory vs. 127438 MiB of total host memory
+0.01.009.512 W llama_context: n_ctx_seq (512) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.01.026.428 I 
+0.01.026.542 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.01.026.548 I compute_imatrix: tokenizing the input ..
+0.01.115.784 I compute_imatrix: tokenization took 89.228 ms
+0.01.115.820 I compute_imatrix: computing over 129 chunks, n_ctx=512, batch_size=512, n_seq=1
+0.05.010.249 I compute_imatrix: 3.89 seconds per pass - ETA 8.37 minutes
+[1]4.8280,[2]3.5310,[3]3.3565,[4]3.5543,[5]3.5306,[6]3.2674,[7]3.7535,[8]3.7744,[9]4.1909,0.35.054.817 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+0.35.054.819 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+0.35.054.834 I 
+0.35.054.839 W save_imatrix: entry '               blk.36.ffn_up_exps.weight' has partial data (99.61%)
+0.35.054.845 W save_imatrix: entry '             blk.34.ffn_down_exps.weight' has partial data (98.83%)
+0.35.054.846 W save_imatrix: entry '               blk.34.ffn_up_exps.weight' has partial data (98.83%)
+0.35.054.849 W save_imatrix: entry '             blk.33.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.852 W save_imatrix: entry '             blk.32.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.856 W save_imatrix: entry '             blk.31.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.857 W save_imatrix: entry '             blk.31.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.860 W save_imatrix: entry '             blk.30.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.864 W save_imatrix: entry '               blk.29.ffn_up_exps.weight' has partial data (98.83%)
+0.35.054.865 W save_imatrix: entry '             blk.29.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.054.874 W save_imatrix: entry '             blk.27.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.877 W save_imatrix: entry '               blk.26.ffn_up_exps.weight' has partial data (99.61%)
+0.35.054.883 W save_imatrix: entry '             blk.24.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.054.893 W save_imatrix: entry '             blk.32.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.895 W save_imatrix: entry '             blk.33.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.897 W save_imatrix: entry '             blk.30.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.903 W save_imatrix: entry '             blk.26.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.907 W save_imatrix: entry '             blk.20.ffn_down_exps.weight' has partial data (98.83%)
+0.35.054.910 W save_imatrix: entry '             blk.22.ffn_down_exps.weight' has partial data (99.22%)
+0.35.054.914 W save_imatrix: entry '             blk.27.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.916 W save_imatrix: entry '             blk.26.ffn_down_exps.weight' has partial data (99.61%)
+0.35.054.919 W save_imatrix: entry '             blk.20.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.054.926 W save_imatrix: entry '               blk.22.ffn_up_exps.weight' has partial data (99.22%)
+0.35.054.928 W save_imatrix: entry '              blk.7.ffn_down_exps.weight' has partial data (99.22%)
+0.35.054.929 W save_imatrix: entry '             blk.36.ffn_gate_exps.weight' has partial data (99.61%)
+0.35.054.933 W save_imatrix: entry '               blk.31.ffn_up_exps.weight' has partial data (99.61%)
+0.35.054.936 W save_imatrix: entry '               blk.20.ffn_up_exps.weight' has partial data (98.83%)
+0.35.054.949 W save_imatrix: entry '             blk.15.ffn_down_exps.weight' has partial data (99.22%)
+0.35.054.968 W save_imatrix: entry '             blk.34.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.054.969 W save_imatrix: entry '              blk.7.ffn_gate_exps.weight' has partial data (99.22%)
+0.35.054.971 W save_imatrix: entry '               blk.33.ffn_up_exps.weight' has partial data (99.61%)
+0.35.054.974 W save_imatrix: entry '             blk.24.ffn_down_exps.weight' has partial data (98.83%)
+0.35.054.983 W save_imatrix: entry '               blk.32.ffn_up_exps.weight' has partial data (99.61%)
+0.35.054.989 W save_imatrix: entry '             blk.18.ffn_down_exps.weight' has partial data (98.83%)
+0.35.054.997 W save_imatrix: entry '             blk.29.ffn_down_exps.weight' has partial data (98.83%)
+0.35.055.001 W save_imatrix: entry '                blk.7.ffn_up_exps.weight' has partial data (99.22%)
+0.35.055.004 W save_imatrix: entry '               blk.18.ffn_up_exps.weight' has partial data (98.83%)
+0.35.055.006 W save_imatrix: entry '               blk.27.ffn_up_exps.weight' has partial data (99.61%)
+0.35.055.013 W save_imatrix: entry '             blk.16.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.055.016 W save_imatrix: entry '               blk.30.ffn_up_exps.weight' has partial data (99.61%)
+0.35.055.019 W save_imatrix: entry '             blk.36.ffn_down_exps.weight' has partial data (99.61%)
+0.35.055.022 W save_imatrix: entry '               blk.16.ffn_up_exps.weight' has partial data (98.83%)
+0.35.055.026 W save_imatrix: entry '             blk.18.ffn_gate_exps.weight' has partial data (98.83%)
+0.35.055.036 W save_imatrix: entry '               blk.24.ffn_up_exps.weight' has partial data (98.83%)
+0.35.055.040 W save_imatrix: entry '             blk.15.ffn_gate_exps.weight' has partial data (99.22%)
+0.35.055.041 W save_imatrix: entry '               blk.15.ffn_up_exps.weight' has partial data (99.22%)
+0.35.055.042 W save_imatrix: entry '             blk.22.ffn_gate_exps.weight' has partial data (99.22%)
+0.35.055.045 W save_imatrix: entry '             blk.16.ffn_down_exps.weight' has partial data (98.83%)
+
+[10]4.3240,[11]4.0211,[12]4.4049,[13]4.9418,[14]5.1664,[15]5.6128,[16]5.8282,[17]6.1242,[18]6.4430,[19]6.1720,1.16.116.747 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+1.16.116.747 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[20]6.1999,[21]6.2244,[22]6.2384,[23]6.2298,[24]6.4258,[25]6.5635,[26]6.5954,[27]6.6758,[28]6.7712,[29]7.0466,1.59.082.255 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+1.59.082.256 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[30]7.0971,[31]6.8928,[32]6.5976,[33]6.3884,[34]6.2660,[35]6.2001,[36]6.3146,[37]6.4046,[38]6.4913,[39]6.6513,2.41.868.430 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+2.41.868.431 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[40]6.8336,[41]6.9385,[42]7.2103,[43]7.3229,[44]7.4938,[45]7.5505,[46]7.5098,[47]7.4647,[48]7.6200,[49]7.6994,3.24.706.480 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+3.24.706.520 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[50]7.6500,[51]7.5877,[52]7.6542,[53]7.7868,[54]7.8872,[55]7.9814,[56]8.0100,[57]8.0181,[58]8.0214,[59]8.0351,4.08.161.216 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+4.08.161.218 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[60]8.0326,[61]7.9840,[62]7.9372,[63]7.9683,[64]8.0068,[65]7.9439,[66]7.9358,[67]7.9336,[68]7.8425,[69]7.8031,5.53.139.568 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+5.53.139.570 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[70]7.8080,[71]7.7716,[72]7.7330,[73]7.7239,[74]7.6582,[75]7.5914,[76]7.5604,[77]7.5344,[78]7.5000,[79]7.4563,6.36.748.709 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+6.36.748.710 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[80]7.3732,[81]7.3905,[82]7.3907,[83]7.3636,[84]7.3800,[85]7.4067,[86]7.3364,[87]7.3262,[88]7.3183,[89]7.3359,7.14.145.562 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+7.14.145.563 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[90]7.3358,[91]7.3166,[92]7.2408,[93]7.1671,[94]7.0855,[95]7.0144,[96]6.9469,[97]6.8749,[98]6.8120,[99]6.7501,7.49.421.037 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+7.49.421.039 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[100]6.7749,[101]6.7937,[102]6.8753,[103]6.9573,[104]7.0285,[105]7.1404,[106]7.2385,[107]7.2672,[108]7.2873,[109]7.2996,8.25.152.163 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+8.25.152.164 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[110]7.3026,[111]7.2653,[112]7.1830,[113]7.1010,[114]7.1408,[115]7.1761,[116]7.1992,[117]7.2053,[118]7.2436,[119]7.2755,9.00.696.257 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+9.00.696.259 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+[120]7.2818,[121]7.2985,[122]7.3085,[123]7.2788,[124]7.3333,[125]7.3907,[126]7.4360,[127]7.5037,[128]7.5475,[129]7.6012,
+Final estimate: PPL = 7.6012 +/- 0.10821
+9.36.441.617 W 
+save_imatrix: saving imatrix using GGUF format with a different suffix than .gguf
+9.36.441.617 W save_imatrix: if you want the previous imatrix format, use --output-format dat
+
+

recipe/logs/N3_readback.log

@@ -0,0 +1,4 @@
+PASS Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_STRIX_LEAN.gguf arch=qwen35moe ftype=106 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q5_K
+PASS Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_COHERENT.gguf arch=qwen35moe ftype=102 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q6_K
+PASS Nex-N2.5-mini-imatrix-Q4_0_ROCMFP4_FAST.gguf arch=qwen35moe ftype=103 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q4_0_ROCMFP4_FAST
+N3_DONE

recipe/logs/N4_kld_q102i.log

@@ -0,0 +1,92 @@
+0.00.037.317 I common_init_result: fitting params to device memory ...
+0.00.037.321 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.390.383 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.21.870.791 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.936.318 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.143.467 I 
+0.22.143.580 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.22.258.917 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.24.621.838 I kl_divergence: 2.36 seconds per pass - ETA 1.57 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.6299 ±    0.4284      -0.01123 ±    0.01762       0.10095 ±    0.00634     9.176 ±  0.605 %    88.270 ±  1.007 %
+   2       6.6587 ±    0.3601      -0.00093 ±    0.01117       0.08359 ±    0.00360     7.978 ±  0.395 %    88.025 ±  0.718 %
+   3       7.1385 ±    0.3226       0.01508 ±    0.00875       0.08242 ±    0.00314     7.696 ±  0.308 %    87.911 ±  0.589 %
+   4       7.4207 ±    0.2990       0.01815 ±    0.00760       0.08463 ±    0.00438     7.777 ±  0.308 %    88.465 ±  0.499 %
+   5       7.2795 ±    0.2642       0.01905 ±    0.00665       0.08239 ±    0.00359     7.768 ±  0.267 %    88.426 ±  0.447 %
+   6       6.2745 ±    0.2014       0.02044 ±    0.00615       0.08353 ±    0.00339     8.576 ±  0.294 %    88.987 ±  0.400 %
+   7       5.8589 ±    0.1719       0.02053 ±    0.00613       0.09196 ±    0.00373     9.221 ±  0.305 %    88.926 ±  0.371 %
+   8       5.7865 ±    0.1577       0.02044 ±    0.00558       0.09026 ±    0.00333     9.081 ±  0.275 %    88.734 ±  0.350 %
+   9       6.0902 ±    0.1574       0.01904 ±    0.00522       0.09017 ±    0.00300     8.885 ±  0.253 %    88.530 ±  0.332 %
+  10       6.2036 ±    0.1532       0.01905 ±    0.00484       0.08664 ±    0.00271     8.652 ±  0.235 %    88.592 ±  0.314 %
+  11       6.2550 ±    0.1467       0.01852 ±    0.00454       0.08389 ±    0.00248     8.471 ±  0.220 %    88.661 ±  0.299 %
+  12       6.5002 ±    0.1472       0.01864 ±    0.00426       0.08096 ±    0.00228     8.271 ±  0.208 %    88.702 ±  0.286 %
+  13       6.5547 ±    0.1424       0.02010 ±    0.00407       0.07947 ±    0.00212     8.179 ±  0.196 %    88.668 ±  0.275 %
+  14       6.6139 ±    0.1383       0.02031 ±    0.00387       0.07794 ±    0.00199     8.054 ±  0.186 %    88.773 ±  0.264 %
+  15       6.6603 ±    0.1347       0.02109 ±    0.00372       0.07724 ±    0.00186     7.982 ±  0.177 %    88.739 ±  0.255 %
+  16       6.8133 ±    0.1336       0.01708 ±    0.00360       0.07665 ±    0.00178     7.924 ±  0.170 %    88.612 ±  0.248 %
+  17       6.8557 ±    0.1300       0.01631 ±    0.00347       0.07576 ±    0.00169     7.851 ±  0.164 %    88.615 ±  0.241 %
+  18       6.9493 ±    0.1282       0.01668 ±    0.00337       0.07548 ±    0.00160     7.811 ±  0.158 %    88.541 ±  0.235 %
+  19       6.9014 ±    0.1244       0.01668 ±    0.00325       0.07445 ±    0.00154     7.725 ±  0.153 %    88.645 ±  0.228 %
+  20       6.6526 ±    0.1161       0.01930 ±    0.00323       0.07848 ±    0.00152     7.994 ±  0.146 %    88.514 ±  0.223 %
+  21       6.6703 ±    0.1135       0.01945 ±    0.00316       0.07876 ±    0.00146     7.982 ±  0.141 %    88.423 ±  0.218 %
+  22       6.6837 ±    0.1111       0.01946 ±    0.00308       0.07952 ±    0.00143     8.019 ±  0.138 %    88.376 ±  0.214 %
+  23       6.7415 ±    0.1096       0.02066 ±    0.00300       0.07952 ±    0.00138     7.991 ±  0.134 %    88.278 ±  0.210 %
+  24       6.7373 ±    0.1070       0.02000 ±    0.00295       0.07956 ±    0.00138     7.986 ±  0.133 %    88.270 ±  0.205 %
+  25       6.7719 ±    0.1054       0.02041 ±    0.00289       0.07925 ±    0.00134     7.953 ±  0.130 %    88.258 ±  0.201 %
+  26       6.7420 ±    0.1028       0.02046 ±    0.00283       0.07951 ±    0.00131     7.991 ±  0.129 %    88.274 ±  0.197 %
+  27       6.9113 ±    0.1041       0.02093 ±    0.00283       0.07927 ±    0.00136     7.933 ±  0.126 %    88.248 ±  0.194 %
+  28       7.0022 ±    0.1039       0.02114 ±    0.00276       0.07854 ±    0.00132     7.873 ±  0.123 %    88.256 ±  0.190 %
+  29       6.9981 ±    0.1020       0.02141 ±    0.00271       0.07833 ±    0.00128     7.882 ±  0.121 %    88.250 ±  0.187 %
+  30       6.9536 ±    0.0995       0.02303 ±    0.00267       0.07848 ±    0.00125     7.892 ±  0.119 %    88.244 ±  0.184 %
+  31       6.8548 ±    0.0962       0.02323 ±    0.00261       0.07786 ±    0.00122     7.868 ±  0.117 %    88.320 ±  0.180 %
+  32       6.7489 ±    0.0931       0.02193 ±    0.00261       0.08043 ±    0.00142     8.006 ±  0.117 %    88.273 ±  0.178 %
+  33       6.6865 ±    0.0906       0.02252 ±    0.00256       0.07990 ±    0.00139     7.989 ±  0.115 %    88.299 ±  0.175 %
+  34       6.6723 ±    0.0889       0.02256 ±    0.00251       0.07907 ±    0.00135     7.928 ±  0.113 %    88.365 ±  0.172 %
+  35       6.6827 ±    0.0878       0.02231 ±    0.00246       0.07849 ±    0.00131     7.908 ±  0.111 %    88.404 ±  0.169 %
+  36       6.6954 ±    0.0868       0.02172 ±    0.00242       0.07826 ±    0.00128     7.892 ±  0.110 %    88.414 ±  0.167 %
+  37       6.6026 ±    0.0841       0.02157 ±    0.00238       0.07763 ±    0.00125     7.875 ±  0.108 %    88.407 ±  0.165 %
+  38       6.5309 ±    0.0818       0.02118 ±    0.00235       0.07760 ±    0.00123     7.895 ±  0.106 %    88.432 ±  0.162 %
+  39       6.4522 ±    0.0795       0.02138 ±    0.00232       0.07746 ±    0.00121     7.905 ±  0.105 %    88.423 ±  0.160 %
+  40       6.3601 ±    0.0770       0.02082 ±    0.00229       0.07695 ±    0.00118     7.909 ±  0.104 %    88.463 ±  0.158 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.360050 ±   0.077045
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  98.21%
+Mean ln(PPL(Q)/PPL(base))     :   0.020824 ±   0.002287
+Mean PPL(Q)/PPL(base)         :   1.021042 ±   0.002336
+Mean PPL(Q)-PPL(base)         :   0.131071 ±   0.014499
+
+====== KL divergence statistics ======
+Mean    KLD:   0.076947 ±   0.001181
+Maximum KLD:  16.293211
+99.9%   KLD:   2.499913
+99.0%   KLD:   0.685225
+95.0%   KLD:   0.259541
+90.0%   KLD:   0.162843
+Median  KLD:   0.034241
+10.0%   KLD:   0.000500
+ 5.0%   KLD:   0.000134
+ 1.0%   KLD:  -0.000030
+ 0.1%   KLD:  -0.000274
+Minimum KLD:  -0.000654
+
+====== Token probability statistics ======
+Mean    Δp: -0.505 ± 0.039 %
+Maximum Δp: 98.233%
+99.9%   Δp: 50.439%
+99.0%   Δp: 21.309%
+95.0%   Δp:  9.386%
+90.0%   Δp:  5.261%
+75.0%   Δp:  0.918%
+Median  Δp: -0.011%
+25.0%   Δp: -1.620%
+10.0%   Δp: -6.708%
+ 5.0%   Δp: -11.584%
+ 1.0%   Δp: -26.637%
+ 0.1%   Δp: -64.007%
+Minimum Δp: -99.856%
+RMS Δp    :  7.909 ± 0.104 %
+Same top p: 88.463 ± 0.158 %
+

recipe/logs/N4_kld_q103i.log

@@ -0,0 +1,92 @@
+0.00.039.723 I common_init_result: fitting params to device memory ...
+0.00.039.726 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.396.902 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.542.121 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.588.569 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.783.826 I 
+0.20.783.939 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.20.898.218 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.22.619.669 I kl_divergence: 1.72 seconds per pass - ETA 1.13 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.7545 ±    0.4426       0.01066 ±    0.01514       0.10310 ±    0.00565     9.252 ±  0.547 %    86.901 ±  1.055 %
+   2       6.7692 ±    0.3702       0.01553 ±    0.01057       0.08979 ±    0.00354     8.313 ±  0.392 %    87.097 ±  0.741 %
+   3       7.2118 ±    0.3286       0.02530 ±    0.00859       0.09142 ±    0.00371     7.924 ±  0.306 %    86.901 ±  0.609 %
+   4       7.4361 ±    0.3008       0.02022 ±    0.00807       0.09891 ±    0.00569     8.314 ±  0.317 %    87.414 ±  0.519 %
+   5       7.2981 ±    0.2661       0.02160 ±    0.00714       0.09587 ±    0.00463     8.276 ±  0.280 %    87.527 ±  0.462 %
+   6       6.3222 ±    0.2036       0.02801 ±    0.00679       0.09876 ±    0.00434     9.164 ±  0.306 %    87.814 ±  0.418 %
+   7       5.8837 ±    0.1729       0.02475 ±    0.00651       0.10562 ±    0.00468     9.504 ±  0.293 %    87.837 ±  0.386 %
+   8       5.8219 ±    0.1591       0.02653 ±    0.00598       0.10395 ±    0.00416     9.429 ±  0.268 %    87.732 ±  0.363 %
+   9       6.1195 ±    0.1584       0.02383 ±    0.00560       0.10437 ±    0.00375     9.307 ±  0.250 %    87.281 ±  0.347 %
+  10       6.2384 ±    0.1544       0.02466 ±    0.00519       0.10032 ±    0.00340     9.080 ±  0.233 %    87.331 ±  0.329 %
+  11       6.3092 ±    0.1486       0.02715 ±    0.00488       0.09769 ±    0.00312     8.996 ±  0.220 %    87.292 ±  0.314 %
+  12       6.5586 ±    0.1493       0.02759 ±    0.00459       0.09446 ±    0.00286     8.773 ±  0.208 %    87.235 ±  0.301 %
+  13       6.6005 ±    0.1440       0.02706 ±    0.00438       0.09304 ±    0.00266     8.721 ±  0.198 %    87.292 ±  0.289 %
+  14       6.6554 ±    0.1398       0.02657 ±    0.00417       0.09137 ±    0.00249     8.562 ±  0.187 %    87.264 ±  0.279 %
+  15       6.6958 ±    0.1359       0.02640 ±    0.00402       0.09068 ±    0.00234     8.515 ±  0.178 %    87.175 ±  0.270 %
+  16       6.8581 ±    0.1350       0.02363 ±    0.00387       0.08962 ±    0.00221     8.427 ±  0.172 %    87.060 ±  0.262 %
+  17       6.9067 ±    0.1315       0.02373 ±    0.00373       0.08844 ±    0.00209     8.335 ±  0.164 %    87.120 ±  0.254 %
+  18       6.9978 ±    0.1297       0.02365 ±    0.00361       0.08829 ±    0.00199     8.313 ±  0.158 %    87.075 ±  0.247 %
+  19       6.9441 ±    0.1257       0.02285 ±    0.00349       0.08712 ±    0.00190     8.208 ±  0.152 %    87.143 ±  0.240 %
+  20       6.6745 ±    0.1169       0.02258 ±    0.00348       0.09166 ±    0.00187     8.508 ±  0.146 %    87.082 ±  0.234 %
+  21       6.6839 ±    0.1140       0.02149 ±    0.00341       0.09200 ±    0.00180     8.487 ±  0.142 %    87.041 ±  0.229 %
+  22       6.6958 ±    0.1115       0.02127 ±    0.00336       0.09301 ±    0.00182     8.516 ±  0.141 %    87.079 ±  0.224 %
+  23       6.7561 ±    0.1101       0.02283 ±    0.00329       0.09310 ±    0.00178     8.515 ±  0.138 %    86.974 ±  0.219 %
+  24       6.7528 ±    0.1075       0.02230 ±    0.00321       0.09247 ±    0.00172     8.468 ±  0.134 %    87.036 ±  0.214 %
+  25       6.7907 ±    0.1060       0.02318 ±    0.00313       0.09180 ±    0.00165     8.416 ±  0.131 %    87.003 ±  0.210 %
+  26       6.7642 ±    0.1034       0.02375 ±    0.00307       0.09231 ±    0.00162     8.481 ±  0.129 %    87.067 ±  0.206 %
+  27       6.9325 ±    0.1047       0.02399 ±    0.00301       0.09164 ±    0.00159     8.400 ±  0.127 %    87.068 ±  0.202 %
+  28       7.0252 ±    0.1045       0.02441 ±    0.00295       0.09094 ±    0.00155     8.369 ±  0.125 %    87.125 ±  0.198 %
+  29       7.0215 ±    0.1026       0.02475 ±    0.00290       0.09104 ±    0.00151     8.421 ±  0.124 %    87.144 ±  0.194 %
+  30       6.9731 ±    0.1000       0.02583 ±    0.00286       0.09121 ±    0.00147     8.431 ±  0.121 %    87.120 ±  0.191 %
+  31       6.8743 ±    0.0967       0.02607 ±    0.00280       0.09057 ±    0.00143     8.408 ±  0.119 %    87.207 ±  0.188 %
+  32       6.7690 ±    0.0936       0.02490 ±    0.00278       0.09337 ±    0.00160     8.561 ±  0.119 %    87.201 ±  0.185 %
+  33       6.7036 ±    0.0910       0.02507 ±    0.00273       0.09280 ±    0.00156     8.553 ±  0.117 %    87.263 ±  0.181 %
+  34       6.6897 ±    0.0893       0.02516 ±    0.00267       0.09179 ±    0.00151     8.490 ±  0.115 %    87.315 ±  0.178 %
+  35       6.6976 ±    0.0881       0.02453 ±    0.00261       0.09113 ±    0.00148     8.437 ±  0.113 %    87.393 ±  0.175 %
+  36       6.7103 ±    0.0872       0.02395 ±    0.00257       0.09082 ±    0.00144     8.395 ±  0.110 %    87.404 ±  0.173 %
+  37       6.6143 ±    0.0845       0.02333 ±    0.00253       0.08990 ±    0.00141     8.356 ±  0.108 %    87.451 ±  0.170 %
+  38       6.5394 ±    0.0821       0.02247 ±    0.00250       0.08996 ±    0.00138     8.378 ±  0.107 %    87.452 ±  0.168 %
+  39       6.4598 ±    0.0798       0.02255 ±    0.00247       0.08964 ±    0.00135     8.375 ±  0.105 %    87.430 ±  0.166 %
+  40       6.3681 ±    0.0773       0.02208 ±    0.00243       0.08897 ±    0.00132     8.370 ±  0.103 %    87.454 ±  0.164 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.368075 ±   0.077317
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.99%
+Mean ln(PPL(Q)/PPL(base))     :   0.022085 ±   0.002430
+Mean PPL(Q)/PPL(base)         :   1.022330 ±   0.002484
+Mean PPL(Q)-PPL(base)         :   0.139096 ±   0.015431
+
+====== KL divergence statistics ======
+Mean    KLD:   0.088974 ±   0.001318
+Maximum KLD:  14.587295
+99.9%   KLD:   2.825675
+99.0%   KLD:   0.784672
+95.0%   KLD:   0.301152
+90.0%   KLD:   0.188552
+Median  KLD:   0.039738
+10.0%   KLD:   0.000526
+ 5.0%   KLD:   0.000124
+ 1.0%   KLD:  -0.000056
+ 0.1%   KLD:  -0.000307
+Minimum KLD:  -0.000906
+
+====== Token probability statistics ======
+Mean    Δp: -0.439 ± 0.041 %
+Maximum Δp: 97.378%
+99.9%   Δp: 51.015%
+99.0%   Δp: 22.861%
+95.0%   Δp: 10.138%
+90.0%   Δp:  5.884%
+75.0%   Δp:  1.147%
+Median  Δp: -0.001%
+25.0%   Δp: -1.525%
+10.0%   Δp: -6.868%
+ 5.0%   Δp: -12.274%
+ 1.0%   Δp: -29.206%
+ 0.1%   Δp: -67.613%
+Minimum Δp: -98.545%
+RMS Δp    :  8.370 ± 0.103 %
+Same top p: 87.454 ± 0.164 %
+

recipe/logs/N4_kld_q106.log

@@ -0,0 +1,92 @@
+0.00.042.882 I common_init_result: fitting params to device memory ...
+0.00.042.887 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.541.611 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.22.255.822 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.22.359.058 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.643.059 I 
+0.22.643.196 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.22.776.319 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.25.307.079 I kl_divergence: 2.53 seconds per pass - ETA 1.68 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       6.2617 ±    0.5005       0.09513 ±    0.01958       0.13331 ±    0.00723    11.273 ±  0.670 %    85.728 ±  1.094 %
+   2       7.1013 ±    0.3976       0.06343 ±    0.01246       0.11005 ±    0.00427     9.422 ±  0.440 %    86.413 ±  0.758 %
+   3       7.5058 ±    0.3468       0.06527 ±    0.01057       0.11313 ±    0.00423     9.250 ±  0.346 %    86.022 ±  0.626 %
+   4       7.7849 ±    0.3204       0.06605 ±    0.00908       0.11249 ±    0.00383     9.264 ±  0.306 %    86.388 ±  0.536 %
+   5       7.5801 ±    0.2810       0.05952 ±    0.00798       0.11043 ±    0.00322     9.087 ±  0.266 %    86.725 ±  0.474 %
+   6       6.5014 ±    0.2131       0.05596 ±    0.00724       0.11026 ±    0.00320     9.619 ±  0.279 %    87.178 ±  0.427 %
+   7       6.0313 ±    0.1806       0.04953 ±    0.00696       0.12172 ±    0.00436    10.069 ±  0.275 %    87.236 ±  0.394 %
+   8       5.9472 ±    0.1655       0.04783 ±    0.00647       0.12058 ±    0.00391     9.949 ±  0.252 %    87.121 ±  0.370 %
+   9       6.2446 ±    0.1643       0.04408 ±    0.00605       0.12012 ±    0.00353     9.791 ±  0.233 %    86.597 ±  0.355 %
+  10       6.3764 ±    0.1603       0.04654 ±    0.00565       0.11624 ±    0.00321     9.609 ±  0.217 %    86.588 ±  0.337 %
+  11       6.4374 ±    0.1538       0.04726 ±    0.00530       0.11268 ±    0.00294     9.442 ±  0.204 %    86.484 ±  0.322 %
+  12       6.6950 ±    0.1547       0.04816 ±    0.00499       0.10924 ±    0.00271     9.238 ±  0.193 %    86.388 ±  0.310 %
+  13       6.7262 ±    0.1490       0.04592 ±    0.00474       0.10773 ±    0.00253     9.174 ±  0.183 %    86.413 ±  0.297 %
+  14       6.7857 ±    0.1447       0.04595 ±    0.00454       0.10583 ±    0.00237     9.050 ±  0.174 %    86.426 ±  0.286 %
+  15       6.8279 ±    0.1408       0.04594 ±    0.00437       0.10522 ±    0.00224     9.021 ±  0.166 %    86.419 ±  0.277 %
+  16       6.9800 ±    0.1394       0.04125 ±    0.00424       0.10488 ±    0.00213     8.960 ±  0.160 %    86.339 ±  0.268 %
+  17       7.0246 ±    0.1356       0.04064 ±    0.00408       0.10348 ±    0.00202     8.863 ±  0.153 %    86.424 ±  0.260 %
+  18       7.1128 ±    0.1335       0.03994 ±    0.00395       0.10297 ±    0.00192     8.819 ±  0.147 %    86.385 ±  0.253 %
+  19       7.0597 ±    0.1294       0.03935 ±    0.00383       0.10135 ±    0.00184     8.726 ±  0.143 %    86.474 ±  0.245 %
+  20       6.7990 ±    0.1207       0.04105 ±    0.00383       0.10693 ±    0.00185     9.139 ±  0.143 %    86.393 ±  0.240 %
+  21       6.8267 ±    0.1182       0.04262 ±    0.00375       0.10765 ±    0.00179     9.144 ±  0.139 %    86.375 ±  0.234 %
+  22       6.8456 ±    0.1159       0.04339 ±    0.00369       0.10893 ±    0.00181     9.184 ±  0.137 %    86.417 ±  0.228 %
+  23       6.9000 ±    0.1143       0.04390 ±    0.00360       0.10839 ±    0.00175     9.147 ±  0.133 %    86.383 ±  0.224 %
+  24       6.8942 ±    0.1115       0.04302 ±    0.00351       0.10851 ±    0.00171     9.165 ±  0.132 %    86.376 ±  0.219 %
+  25       6.9297 ±    0.1099       0.04345 ±    0.00344       0.10819 ±    0.00165     9.128 ±  0.128 %    86.334 ±  0.215 %
+  26       6.8989 ±    0.1072       0.04347 ±    0.00338       0.10899 ±    0.00166     9.185 ±  0.127 %    86.371 ±  0.210 %
+  27       7.0668 ±    0.1084       0.04318 ±    0.00330       0.10805 ±    0.00160     9.103 ±  0.124 %    86.384 ±  0.206 %
+  28       7.1505 ±    0.1080       0.04210 ±    0.00321       0.10685 ±    0.00155     9.028 ±  0.122 %    86.395 ±  0.203 %
+  29       7.1560 ±    0.1063       0.04372 ±    0.00317       0.10671 ±    0.00152     9.048 ±  0.120 %    86.392 ±  0.199 %
+  30       7.0985 ±    0.1034       0.04365 ±    0.00312       0.10649 ±    0.00148     9.044 ±  0.118 %    86.341 ±  0.196 %
+  31       6.9900 ±    0.0998       0.04276 ±    0.00305       0.10563 ±    0.00144     9.018 ±  0.116 %    86.444 ±  0.192 %
+  32       6.8747 ±    0.0965       0.04039 ±    0.00305       0.10917 ±    0.00170     9.201 ±  0.119 %    86.416 ±  0.189 %
+  33       6.8130 ±    0.0939       0.04125 ±    0.00301       0.10876 ±    0.00166     9.199 ±  0.117 %    86.457 ±  0.186 %
+  34       6.7927 ±    0.0920       0.04044 ±    0.00294       0.10768 ±    0.00161     9.128 ±  0.115 %    86.441 ±  0.184 %
+  35       6.8056 ±    0.0909       0.04054 ±    0.00288       0.10692 ±    0.00157     9.083 ±  0.113 %    86.494 ±  0.181 %
+  36       6.8222 ±    0.0899       0.04049 ±    0.00284       0.10659 ±    0.00154     9.060 ±  0.111 %    86.546 ±  0.178 %
+  37       6.7263 ±    0.0872       0.04013 ±    0.00279       0.10571 ±    0.00150     9.033 ±  0.109 %    86.566 ±  0.175 %
+  38       6.6472 ±    0.0847       0.03882 ±    0.00275       0.10525 ±    0.00147     9.020 ±  0.107 %    86.621 ±  0.173 %
+  39       6.5675 ±    0.0823       0.03909 ±    0.00271       0.10501 ±    0.00144     9.022 ±  0.106 %    86.638 ±  0.170 %
+  40       6.4740 ±    0.0798       0.03858 ±    0.00267       0.10444 ±    0.00141     9.028 ±  0.105 %    86.659 ±  0.168 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.473964 ±   0.079799
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.63%
+Mean ln(PPL(Q)/PPL(base))     :   0.038576 ±   0.002667
+Mean PPL(Q)/PPL(base)         :   1.039330 ±   0.002772
+Mean PPL(Q)-PPL(base)         :   0.244985 ±   0.017453
+
+====== KL divergence statistics ======
+Mean    KLD:   0.104436 ±   0.001413
+Maximum KLD:  13.909879
+99.9%   KLD:   3.189192
+99.0%   KLD:   0.926687
+95.0%   KLD:   0.354656
+90.0%   KLD:   0.225739
+Median  KLD:   0.047702
+10.0%   KLD:   0.000698
+ 5.0%   KLD:   0.000183
+ 1.0%   KLD:  -0.000033
+ 0.1%   KLD:  -0.000321
+Minimum KLD:  -0.000718
+
+====== Token probability statistics ======
+Mean    Δp: -0.244 ± 0.045 %
+Maximum Δp: 99.706%
+99.9%   Δp: 59.493%
+99.0%   Δp: 25.100%
+95.0%   Δp: 11.707%
+90.0%   Δp:  6.963%
+75.0%   Δp:  1.453%
+Median  Δp: -0.001%
+25.0%   Δp: -1.479%
+10.0%   Δp: -7.240%
+ 5.0%   Δp: -12.992%
+ 1.0%   Δp: -31.268%
+ 0.1%   Δp: -68.467%
+Minimum Δp: -98.412%
+RMS Δp    :  9.028 ± 0.105 %
+Same top p: 86.659 ± 0.168 %
+

recipe/logs/N4_kld_q106i.log

@@ -0,0 +1,92 @@
+0.00.061.730 I common_init_result: fitting params to device memory ...
+0.00.061.734 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.418.389 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.21.286.189 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.331.409 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.21.529.206 I 
+0.21.529.313 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.21.655.658 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.23.558.651 I kl_divergence: 1.90 seconds per pass - ETA 1.27 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.7279 ±    0.4388       0.00602 ±    0.01614       0.10315 ±    0.00583     9.735 ±  0.571 %    85.924 ±  1.088 %
+   2       6.7923 ±    0.3712       0.01894 ±    0.01065       0.08743 ±    0.00335     8.343 ±  0.375 %    86.413 ±  0.758 %
+   3       7.2151 ±    0.3282       0.02577 ±    0.00843       0.08904 ±    0.00386     8.063 ±  0.325 %    86.999 ±  0.607 %
+   4       7.4647 ±    0.3021       0.02405 ±    0.00738       0.09108 ±    0.00535     8.270 ±  0.318 %    87.634 ±  0.515 %
+   5       7.3108 ±    0.2664       0.02334 ±    0.00659       0.08904 ±    0.00437     8.208 ±  0.278 %    87.801 ±  0.458 %
+   6       6.2917 ±    0.2025       0.02318 ±    0.00607       0.08932 ±    0.00383     8.727 ±  0.274 %    88.237 ±  0.411 %
+   7       5.8298 ±    0.1710       0.01556 ±    0.00599       0.09705 ±    0.00425     9.211 ±  0.278 %    88.242 ±  0.381 %
+   8       5.7710 ±    0.1572       0.01775 ±    0.00555       0.09614 ±    0.00379     9.187 ±  0.255 %    88.038 ±  0.359 %
+   9       6.0701 ±    0.1567       0.01574 ±    0.00523       0.09667 ±    0.00342     9.059 ±  0.236 %    87.553 ±  0.344 %
+  10       6.1910 ±    0.1528       0.01702 ±    0.00487       0.09332 ±    0.00310     8.829 ±  0.220 %    87.625 ±  0.326 %
+  11       6.2600 ±    0.1469       0.01931 ±    0.00460       0.09110 ±    0.00284     8.756 ±  0.210 %    87.630 ±  0.310 %
+  12       6.5141 ±    0.1478       0.02077 ±    0.00433       0.08827 ±    0.00261     8.573 ±  0.198 %    87.569 ±  0.298 %
+  13       6.5510 ±    0.1424       0.01954 ±    0.00414       0.08704 ±    0.00243     8.533 ±  0.188 %    87.533 ±  0.286 %
+  14       6.6159 ±    0.1386       0.02062 ±    0.00397       0.08562 ±    0.00227     8.395 ±  0.178 %    87.565 ±  0.276 %
+  15       6.6597 ±    0.1349       0.02101 ±    0.00382       0.08498 ±    0.00213     8.353 ±  0.170 %    87.481 ±  0.267 %
+  16       6.8126 ±    0.1337       0.01697 ±    0.00370       0.08434 ±    0.00203     8.278 ±  0.164 %    87.390 ±  0.259 %
+  17       6.8618 ±    0.1303       0.01721 ±    0.00357       0.08322 ±    0.00192     8.183 ±  0.157 %    87.511 ±  0.251 %
+  18       6.9534 ±    0.1285       0.01728 ±    0.00345       0.08273 ±    0.00182     8.130 ±  0.150 %    87.471 ±  0.244 %
+  19       6.9010 ±    0.1245       0.01662 ±    0.00334       0.08204 ±    0.00175     8.059 ±  0.147 %    87.508 ±  0.237 %
+  20       6.6434 ±    0.1160       0.01790 ±    0.00333       0.08651 ±    0.00173     8.357 ±  0.142 %    87.454 ±  0.232 %
+  21       6.6543 ±    0.1132       0.01704 ±    0.00325       0.08675 ±    0.00166     8.313 ±  0.137 %    87.399 ±  0.226 %
+  22       6.6714 ±    0.1109       0.01762 ±    0.00322       0.08786 ±    0.00171     8.337 ±  0.136 %    87.412 ±  0.221 %
+  23       6.7286 ±    0.1095       0.01874 ±    0.00316       0.08790 ±    0.00167     8.321 ±  0.134 %    87.390 ±  0.216 %
+  24       6.7323 ±    0.1070       0.01926 ±    0.00310       0.08781 ±    0.00164     8.305 ±  0.133 %    87.447 ±  0.211 %
+  25       6.7692 ±    0.1055       0.02001 ±    0.00302       0.08721 ±    0.00158     8.253 ±  0.129 %    87.457 ±  0.207 %
+  26       6.7420 ±    0.1029       0.02046 ±    0.00296       0.08758 ±    0.00155     8.276 ±  0.128 %    87.522 ±  0.203 %
+  27       6.9087 ±    0.1041       0.02056 ±    0.00290       0.08699 ±    0.00151     8.214 ±  0.125 %    87.520 ±  0.199 %
+  28       6.9975 ±    0.1039       0.02047 ±    0.00285       0.08640 ±    0.00147     8.197 ±  0.124 %    87.558 ±  0.195 %
+  29       6.9967 ±    0.1020       0.02121 ±    0.00280       0.08643 ±    0.00144     8.249 ±  0.123 %    87.538 ±  0.192 %
+  30       6.9524 ±    0.0996       0.02286 ±    0.00276       0.08671 ±    0.00140     8.260 ±  0.120 %    87.511 ±  0.189 %
+  31       6.8530 ±    0.0962       0.02296 ±    0.00270       0.08616 ±    0.00136     8.244 ±  0.118 %    87.623 ±  0.185 %
+  32       6.7437 ±    0.0931       0.02116 ±    0.00271       0.08891 ±    0.00158     8.411 ±  0.120 %    87.628 ±  0.182 %
+  33       6.6801 ±    0.0905       0.02156 ±    0.00266       0.08849 ±    0.00154     8.418 ±  0.119 %    87.666 ±  0.179 %
+  34       6.6705 ±    0.0889       0.02230 ±    0.00261       0.08778 ±    0.00150     8.378 ±  0.117 %    87.715 ±  0.176 %
+  35       6.6811 ±    0.0878       0.02207 ±    0.00255       0.08711 ±    0.00147     8.332 ±  0.115 %    87.792 ±  0.173 %
+  36       6.6976 ±    0.0869       0.02206 ±    0.00252       0.08700 ±    0.00143     8.296 ±  0.113 %    87.814 ±  0.170 %
+  37       6.6017 ±    0.0842       0.02143 ±    0.00248       0.08616 ±    0.00140     8.262 ±  0.111 %    87.823 ±  0.168 %
+  38       6.5268 ±    0.0818       0.02054 ±    0.00245       0.08605 ±    0.00137     8.265 ±  0.109 %    87.853 ±  0.166 %
+  39       6.4448 ±    0.0795       0.02022 ±    0.00241       0.08583 ±    0.00134     8.292 ±  0.107 %    87.829 ±  0.164 %
+  40       6.3536 ±    0.0770       0.01981 ±    0.00238       0.08516 ±    0.00131     8.281 ±  0.106 %    87.849 ±  0.162 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.353616 ±   0.077001
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  98.08%
+Mean ln(PPL(Q)/PPL(base))     :   0.019812 ±   0.002375
+Mean PPL(Q)/PPL(base)         :   1.020009 ±   0.002423
+Mean PPL(Q)-PPL(base)         :   0.124637 ±   0.015034
+
+====== KL divergence statistics ======
+Mean    KLD:   0.085158 ±   0.001309
+Maximum KLD:  16.350994
+99.9%   KLD:   2.569785
+99.0%   KLD:   0.737186
+95.0%   KLD:   0.287234
+90.0%   KLD:   0.182120
+Median  KLD:   0.038436
+10.0%   KLD:   0.000529
+ 5.0%   KLD:   0.000132
+ 1.0%   KLD:  -0.000040
+ 0.1%   KLD:  -0.000280
+Minimum KLD:  -0.000917
+
+====== Token probability statistics ======
+Mean    Δp: -0.453 ± 0.041 %
+Maximum Δp: 99.545%
+99.9%   Δp: 55.136%
+99.0%   Δp: 23.249%
+95.0%   Δp:  9.945%
+90.0%   Δp:  5.745%
+75.0%   Δp:  0.987%
+Median  Δp: -0.006%
+25.0%   Δp: -1.591%
+10.0%   Δp: -7.009%
+ 5.0%   Δp: -11.986%
+ 1.0%   Δp: -27.836%
+ 0.1%   Δp: -65.986%
+Minimum Δp: -99.892%
+RMS Δp    :  8.281 ± 0.106 %
+Same top p: 87.849 ± 0.162 %
+

recipe/logs/N4v_kld_q102i.log

@@ -0,0 +1,93 @@
+0.00.035.061 I common_init_result: fitting params to device memory ...
+0.00.035.064 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.364.958 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.365.200 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.20.724.587 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.776.655 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.958.449 I 
+0.20.958.535 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.21.072.467 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.23.478.059 I kl_divergence: 2.41 seconds per pass - ETA 1.60 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.5867 ±    0.4228      -0.01894 ±    0.01754       0.10256 ±    0.00648     9.538 ±  0.610 %    88.368 ±  1.003 %
+   2       6.6450 ±    0.3596      -0.00300 ±    0.01113       0.08488 ±    0.00372     8.059 ±  0.404 %    88.270 ±  0.712 %
+   3       7.0875 ±    0.3197       0.00792 ±    0.00873       0.08299 ±    0.00332     7.716 ±  0.310 %    88.172 ±  0.583 %
+   4       7.3455 ±    0.2950       0.00796 ±    0.00746       0.08559 ±    0.00479     7.862 ±  0.310 %    88.759 ±  0.494 %
+   5       7.2194 ±    0.2615       0.01075 ±    0.00655       0.08256 ±    0.00391     7.727 ±  0.267 %    88.524 ±  0.446 %
+   6       6.2395 ±    0.1998       0.01485 ±    0.00603       0.08240 ±    0.00357     8.402 ±  0.286 %    88.970 ±  0.400 %
+   7       5.8467 ±    0.1714       0.01845 ±    0.00617       0.09250 ±    0.00438     9.020 ±  0.301 %    88.884 ±  0.371 %
+   8       5.7766 ±    0.1573       0.01872 ±    0.00560       0.09000 ±    0.00387     8.841 ±  0.272 %    88.869 ±  0.348 %
+   9       6.0753 ±    0.1568       0.01659 ±    0.00524       0.08968 ±    0.00347     8.651 ±  0.250 %    88.617 ±  0.331 %
+  10       6.1938 ±    0.1528       0.01748 ±    0.00486       0.08628 ±    0.00314     8.443 ±  0.232 %    88.778 ±  0.312 %
+  11       6.2507 ±    0.1465       0.01783 ±    0.00456       0.08369 ±    0.00287     8.284 ±  0.217 %    88.812 ±  0.297 %
+  12       6.4984 ±    0.1471       0.01836 ±    0.00428       0.08087 ±    0.00264     8.100 ±  0.205 %    88.840 ±  0.284 %
+  13       6.5527 ±    0.1424       0.01980 ±    0.00408       0.07955 ±    0.00245     8.022 ±  0.193 %    88.646 ±  0.275 %
+  14       6.6129 ±    0.1383       0.02016 ±    0.00389       0.07798 ±    0.00228     7.902 ±  0.183 %    88.752 ±  0.264 %
+  15       6.6619 ±    0.1348       0.02133 ±    0.00373       0.07730 ±    0.00214     7.842 ±  0.174 %    88.693 ±  0.256 %
+  16       6.8209 ±    0.1338       0.01819 ±    0.00361       0.07652 ±    0.00203     7.773 ±  0.167 %    88.618 ±  0.248 %
+  17       6.8676 ±    0.1304       0.01805 ±    0.00348       0.07551 ±    0.00192     7.696 ±  0.162 %    88.609 ±  0.241 %
+  18       6.9618 ±    0.1286       0.01849 ±    0.00338       0.07510 ±    0.00182     7.653 ±  0.155 %    88.552 ±  0.235 %
+  19       6.9170 ±    0.1249       0.01894 ±    0.00325       0.07404 ±    0.00175     7.561 ±  0.151 %    88.661 ±  0.227 %
+  20       6.6681 ±    0.1166       0.02161 ±    0.00323       0.07796 ±    0.00171     7.836 ±  0.145 %    88.548 ±  0.223 %
+  21       6.6815 ±    0.1138       0.02112 ±    0.00316       0.07831 ±    0.00165     7.820 ±  0.139 %    88.428 ±  0.218 %
+  22       6.6992 ±    0.1116       0.02177 ±    0.00313       0.07958 ±    0.00169     7.862 ±  0.139 %    88.394 ±  0.214 %
+  23       6.7561 ±    0.1101       0.02283 ±    0.00306       0.07988 ±    0.00164     7.864 ±  0.136 %    88.270 ±  0.210 %
+  24       6.7528 ±    0.1074       0.02230 ±    0.00301       0.07975 ±    0.00161     7.848 ±  0.134 %    88.262 ±  0.205 %
+  25       6.7834 ±    0.1058       0.02210 ±    0.00294       0.07935 ±    0.00155     7.808 ±  0.131 %    88.223 ±  0.202 %
+  26       6.7551 ±    0.1032       0.02241 ±    0.00288       0.07966 ±    0.00152     7.846 ±  0.129 %    88.270 ±  0.197 %
+  27       6.9239 ±    0.1045       0.02275 ±    0.00288       0.07942 ±    0.00156     7.792 ±  0.127 %    88.244 ±  0.194 %
+  28       7.0172 ±    0.1043       0.02329 ±    0.00282       0.07891 ±    0.00153     7.749 ±  0.124 %    88.259 ±  0.190 %
+  29       7.0127 ±    0.1024       0.02349 ±    0.00277       0.07872 ±    0.00148     7.761 ±  0.122 %    88.263 ±  0.187 %
+  30       6.9664 ±    0.0999       0.02487 ±    0.00272       0.07871 ±    0.00144     7.761 ±  0.119 %    88.263 ±  0.184 %
+  31       6.8666 ±    0.0966       0.02496 ±    0.00265       0.07794 ±    0.00140     7.729 ±  0.117 %    88.377 ±  0.180 %
+  32       6.7575 ±    0.0934       0.02320 ±    0.00266       0.08099 ±    0.00163     7.910 ±  0.120 %    88.313 ±  0.178 %
+  33       6.6952 ±    0.0909       0.02382 ±    0.00260       0.08035 ±    0.00159     7.884 ±  0.118 %    88.371 ±  0.174 %
+  34       6.6814 ±    0.0892       0.02392 ±    0.00254       0.07951 ±    0.00154     7.831 ±  0.115 %    88.416 ±  0.172 %
+  35       6.6908 ±    0.0880       0.02351 ±    0.00249       0.07890 ±    0.00150     7.807 ±  0.113 %    88.465 ±  0.169 %
+  36       6.7037 ±    0.0870       0.02296 ±    0.00245       0.07859 ±    0.00146     7.786 ±  0.111 %    88.498 ±  0.166 %
+  37       6.6096 ±    0.0844       0.02263 ±    0.00241       0.07791 ±    0.00143     7.765 ±  0.109 %    88.492 ±  0.164 %
+  38       6.5383 ±    0.0821       0.02230 ±    0.00238       0.07778 ±    0.00140     7.790 ±  0.108 %    88.514 ±  0.162 %
+  39       6.4567 ±    0.0797       0.02207 ±    0.00234       0.07744 ±    0.00136     7.798 ±  0.106 %    88.508 ±  0.160 %
+  40       6.3632 ±    0.0772       0.02132 ±    0.00231       0.07684 ±    0.00133     7.783 ±  0.104 %    88.556 ±  0.157 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.363210 ±   0.077202
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  98.19%
+Mean ln(PPL(Q)/PPL(base))     :   0.021321 ±   0.002307
+Mean PPL(Q)/PPL(base)         :   1.021549 ±   0.002357
+Mean PPL(Q)-PPL(base)         :   0.134231 ±   0.014643
+
+====== KL divergence statistics ======
+Mean    KLD:   0.076844 ±   0.001332
+Maximum KLD:  16.142012
+99.9%   KLD:   2.816024
+99.0%   KLD:   0.669352
+95.0%   KLD:   0.255460
+90.0%   KLD:   0.161866
+Median  KLD:   0.033844
+10.0%   KLD:   0.000495
+ 5.0%   KLD:   0.000131
+ 1.0%   KLD:  -0.000030
+ 0.1%   KLD:  -0.000287
+Minimum KLD:  -0.000637
+
+====== Token probability statistics ======
+Mean    Δp: -0.458 ± 0.038 %
+Maximum Δp: 99.702%
+99.9%   Δp: 52.654%
+99.0%   Δp: 20.917%
+95.0%   Δp:  9.373%
+90.0%   Δp:  5.248%
+75.0%   Δp:  0.940%
+Median  Δp: -0.010%
+25.0%   Δp: -1.540%
+10.0%   Δp: -6.629%
+ 5.0%   Δp: -11.288%
+ 1.0%   Δp: -25.973%
+ 0.1%   Δp: -60.301%
+Minimum Δp: -99.838%
+RMS Δp    :  7.783 ± 0.104 %
+Same top p: 88.556 ± 0.157 %
+

recipe/logs/N4v_kld_q103.log

@@ -0,0 +1,93 @@
+0.00.042.837 I common_init_result: fitting params to device memory ...
+0.00.042.842 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.444.710 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.459.450 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.21.377.631 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.428.701 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.21.707.133 I 
+0.21.707.313 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.21.843.964 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.24.857.762 I kl_divergence: 3.01 seconds per pass - ETA 2.00 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       6.1749 ±    0.4909       0.08117 ±    0.01859       0.12763 ±    0.00703    10.392 ±  0.610 %    85.337 ±  1.107 %
+   2       7.1137 ±    0.3991       0.06516 ±    0.01199       0.10730 ±    0.00416     9.008 ±  0.411 %    86.266 ±  0.761 %
+   3       7.5580 ±    0.3515       0.07219 ±    0.00958       0.10875 ±    0.00396     8.609 ±  0.316 %    85.989 ±  0.627 %
+   4       7.8280 ±    0.3239       0.07158 ±    0.00898       0.12059 ±    0.00615     9.140 ±  0.324 %    86.193 ±  0.539 %
+   5       7.6146 ±    0.2833       0.06405 ±    0.00801       0.11743 ±    0.00502     9.043 ±  0.279 %    86.549 ±  0.477 %
+   6       6.5667 ±    0.2161       0.06595 ±    0.00747       0.11888 ±    0.00464     9.905 ±  0.299 %    86.804 ±  0.432 %
+   7       6.0824 ±    0.1827       0.05798 ±    0.00709       0.12603 ±    0.00484    10.284 ±  0.293 %    86.929 ±  0.398 %
+   8       5.9850 ±    0.1670       0.05416 ±    0.00651       0.12399 ±    0.00430    10.146 ±  0.267 %    86.877 ±  0.373 %
+   9       6.2982 ±    0.1664       0.05262 ±    0.00612       0.12384 ±    0.00387     9.948 ±  0.246 %    86.445 ±  0.357 %
+  10       6.4302 ±    0.1624       0.05493 ±    0.00573       0.12014 ±    0.00351     9.781 ±  0.229 %    86.452 ±  0.338 %
+  11       6.4862 ±    0.1556       0.05482 ±    0.00537       0.11666 ±    0.00323     9.576 ±  0.215 %    86.564 ±  0.322 %
+  12       6.7484 ±    0.1564       0.05611 ±    0.00505       0.11329 ±    0.00297     9.389 ±  0.203 %    86.470 ±  0.309 %
+  13       6.7729 ±    0.1504       0.05285 ±    0.00482       0.11177 ±    0.00277     9.295 ±  0.192 %    86.525 ±  0.296 %
+  14       6.8361 ±    0.1461       0.05335 ±    0.00462       0.11058 ±    0.00260     9.192 ±  0.183 %    86.385 ±  0.287 %
+  15       6.8710 ±    0.1419       0.05223 ±    0.00443       0.11010 ±    0.00248     9.182 ±  0.175 %    86.373 ±  0.277 %
+  16       7.0369 ±    0.1408       0.04937 ±    0.00428       0.10941 ±    0.00236     9.094 ±  0.169 %    86.278 ±  0.269 %
+  17       7.0798 ±    0.1370       0.04848 ±    0.00412       0.10796 ±    0.00224     9.005 ±  0.162 %    86.361 ±  0.260 %
+  18       7.1674 ±    0.1348       0.04759 ±    0.00399       0.10751 ±    0.00213     8.990 ±  0.157 %    86.201 ±  0.254 %
+  19       7.1262 ±    0.1310       0.04873 ±    0.00389       0.10651 ±    0.00204     8.894 ±  0.152 %    86.227 ±  0.247 %
+  20       6.8634 ±    0.1222       0.05049 ±    0.00387       0.11186 ±    0.00203     9.240 ±  0.148 %    86.202 ±  0.241 %
+  21       6.8908 ±    0.1196       0.05197 ±    0.00379       0.11250 ±    0.00196     9.228 ±  0.144 %    86.175 ±  0.235 %
+  22       6.9164 ±    0.1174       0.05368 ±    0.00373       0.11404 ±    0.00194     9.255 ±  0.141 %    86.164 ±  0.230 %
+  23       6.9778 ±    0.1159       0.05511 ±    0.00365       0.11379 ±    0.00188     9.234 ±  0.139 %    86.128 ±  0.225 %
+  24       6.9773 ±    0.1133       0.05500 ±    0.00356       0.11399 ±    0.00184     9.243 ±  0.137 %    86.038 ±  0.221 %
+  25       7.0132 ±    0.1116       0.05542 ±    0.00348       0.11365 ±    0.00178     9.211 ±  0.133 %    86.037 ±  0.217 %
+  26       6.9824 ±    0.1089       0.05549 ±    0.00342       0.11396 ±    0.00176     9.257 ±  0.131 %    86.070 ±  0.212 %
+  27       7.1525 ±    0.1102       0.05524 ±    0.00337       0.11346 ±    0.00174     9.179 ±  0.129 %    86.072 ±  0.208 %
+  28       7.2389 ±    0.1098       0.05439 ±    0.00329       0.11219 ±    0.00168     9.107 ±  0.126 %    86.084 ±  0.205 %
+  29       7.2404 ±    0.1079       0.05545 ±    0.00323       0.11206 ±    0.00164     9.137 ±  0.125 %    86.048 ±  0.201 %
+  30       7.1812 ±    0.1050       0.05523 ±    0.00318       0.11167 ±    0.00159     9.127 ±  0.122 %    86.022 ±  0.198 %
+  31       7.0772 ±    0.1015       0.05516 ±    0.00311       0.11097 ±    0.00155     9.104 ±  0.119 %    86.151 ±  0.194 %
+  32       6.9648 ±    0.0981       0.05342 ±    0.00309       0.11357 ±    0.00169     9.252 ±  0.121 %    86.092 ±  0.191 %
+  33       6.9013 ±    0.0955       0.05413 ±    0.00306       0.11324 ±    0.00166     9.238 ±  0.119 %    86.087 ±  0.188 %
+  34       6.8839 ±    0.0937       0.05378 ±    0.00300       0.11241 ±    0.00162     9.209 ±  0.119 %    86.093 ±  0.186 %
+  35       6.8905 ±    0.0924       0.05293 ±    0.00293       0.11139 ±    0.00158     9.152 ±  0.116 %    86.178 ±  0.182 %
+  36       6.9081 ±    0.0914       0.05301 ±    0.00290       0.11111 ±    0.00155     9.127 ±  0.114 %    86.190 ±  0.180 %
+  37       6.8115 ±    0.0887       0.05270 ±    0.00284       0.11014 ±    0.00151     9.096 ±  0.112 %    86.238 ±  0.177 %
+  38       6.7312 ±    0.0861       0.05138 ±    0.00280       0.10982 ±    0.00148     9.100 ±  0.110 %    86.299 ±  0.174 %
+  39       6.6496 ±    0.0837       0.05151 ±    0.00276       0.10945 ±    0.00145     9.096 ±  0.108 %    86.312 ±  0.172 %
+  40       6.5513 ±    0.0811       0.05045 ±    0.00272       0.10876 ±    0.00142     9.101 ±  0.107 %    86.356 ±  0.170 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.551305 ±   0.081055
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.55%
+Mean ln(PPL(Q)/PPL(base))     :   0.050452 ±   0.002720
+Mean PPL(Q)/PPL(base)         :   1.051746 ±   0.002861
+Mean PPL(Q)-PPL(base)         :   0.322326 ±   0.018212
+
+====== KL divergence statistics ======
+Mean    KLD:   0.108756 ±   0.001422
+Maximum KLD:  12.410495
+99.9%   KLD:   3.419987
+99.0%   KLD:   0.930024
+95.0%   KLD:   0.366912
+90.0%   KLD:   0.235489
+Median  KLD:   0.050197
+10.0%   KLD:   0.000729
+ 5.0%   KLD:   0.000189
+ 1.0%   KLD:  -0.000051
+ 0.1%   KLD:  -0.000341
+Minimum KLD:  -0.000842
+
+====== Token probability statistics ======
+Mean    Δp: -0.471 ± 0.045 %
+Maximum Δp: 96.576%
+99.9%   Δp: 54.353%
+99.0%   Δp: 24.394%
+95.0%   Δp: 11.471%
+90.0%   Δp:  6.631%
+75.0%   Δp:  1.298%
+Median  Δp: -0.006%
+25.0%   Δp: -1.652%
+10.0%   Δp: -7.598%
+ 5.0%   Δp: -13.545%
+ 1.0%   Δp: -32.141%
+ 0.1%   Δp: -70.328%
+Minimum Δp: -99.714%
+RMS Δp    :  9.101 ± 0.107 %
+Same top p: 86.356 ± 0.170 %
+

recipe/logs/N4v_kld_q103i.log

@@ -0,0 +1,93 @@
+0.00.033.949 I common_init_result: fitting params to device memory ...
+0.00.033.953 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.371.029 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.312.258 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.19.504.583 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.19.553.410 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.19.731.487 I 
+0.19.731.587 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.19.847.243 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.22.241.158 I kl_divergence: 2.39 seconds per pass - ETA 1.58 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.7501 ±    0.4413       0.00989 ±    0.01572       0.10954 ±    0.00645     9.805 ±  0.583 %    85.630 ±  1.097 %
+   2       6.7714 ±    0.3709       0.01585 ±    0.01070       0.09308 ±    0.00377     8.559 ±  0.392 %    86.706 ±  0.751 %
+   3       7.1773 ±    0.3266       0.02051 ±    0.00847       0.09216 ±    0.00368     7.997 ±  0.296 %    86.934 ±  0.608 %
+   4       7.4782 ±    0.3040       0.02587 ±    0.00786       0.09738 ±    0.00472     8.531 ±  0.336 %    87.341 ±  0.520 %
+   5       7.3506 ±    0.2694       0.02877 ±    0.00704       0.09557 ±    0.00398     8.489 ±  0.308 %    87.468 ±  0.463 %
+   6       6.3429 ±    0.2051       0.03128 ±    0.00664       0.09662 ±    0.00373     9.145 ±  0.308 %    87.781 ±  0.418 %
+   7       5.9472 ±    0.1761       0.03550 ±    0.00668       0.10791 ±    0.00462     9.889 ±  0.320 %    87.795 ±  0.387 %
+   8       5.8786 ±    0.1618       0.03622 ±    0.00612       0.10626 ±    0.00411     9.788 ±  0.292 %    87.683 ±  0.363 %
+   9       6.1674 ±    0.1605       0.03164 ±    0.00571       0.10586 ±    0.00370     9.627 ±  0.271 %    87.227 ±  0.348 %
+  10       6.2930 ±    0.1566       0.03336 ±    0.00529       0.10181 ±    0.00335     9.388 ±  0.252 %    87.390 ±  0.328 %
+  11       6.3574 ±    0.1505       0.03475 ±    0.00499       0.09913 ±    0.00308     9.301 ±  0.238 %    87.292 ±  0.314 %
+  12       6.6087 ±    0.1511       0.03519 ±    0.00469       0.09595 ±    0.00284     9.076 ±  0.225 %    87.317 ±  0.300 %
+  13       6.6444 ±    0.1456       0.03370 ±    0.00446       0.09418 ±    0.00263     8.998 ±  0.213 %    87.367 ±  0.288 %
+  14       6.6988 ±    0.1413       0.03307 ±    0.00425       0.09228 ±    0.00246     8.826 ±  0.202 %    87.467 ±  0.277 %
+  15       6.7357 ±    0.1373       0.03235 ±    0.00408       0.09142 ±    0.00231     8.745 ±  0.192 %    87.488 ±  0.267 %
+  16       6.8972 ±    0.1363       0.02932 ±    0.00393       0.09032 ±    0.00219     8.621 ±  0.183 %    87.390 ±  0.259 %
+  17       6.9396 ±    0.1326       0.02848 ±    0.00379       0.08911 ±    0.00207     8.521 ±  0.175 %    87.436 ±  0.251 %
+  18       7.0315 ±    0.1308       0.02845 ±    0.00366       0.08879 ±    0.00197     8.486 ±  0.168 %    87.374 ±  0.245 %
+  19       6.9777 ±    0.1268       0.02767 ±    0.00353       0.08759 ±    0.00188     8.384 ±  0.163 %    87.385 ±  0.238 %
+  20       6.7038 ±    0.1178       0.02695 ±    0.00352       0.09228 ±    0.00186     8.685 ±  0.156 %    87.273 ±  0.233 %
+  21       6.7149 ±    0.1149       0.02612 ±    0.00345       0.09269 ±    0.00179     8.668 ±  0.151 %    87.162 ±  0.228 %
+  22       6.7275 ±    0.1124       0.02599 ±    0.00339       0.09343 ±    0.00179     8.678 ±  0.150 %    87.186 ±  0.223 %
+  23       6.7910 ±    0.1111       0.02798 ±    0.00333       0.09354 ±    0.00176     8.670 ±  0.147 %    87.080 ±  0.219 %
+  24       6.7874 ±    0.1085       0.02741 ±    0.00325       0.09330 ±    0.00171     8.652 ±  0.144 %    87.097 ±  0.214 %
+  25       6.8244 ±    0.1069       0.02813 ±    0.00316       0.09247 ±    0.00164     8.585 ±  0.140 %    87.077 ±  0.210 %
+  26       6.7976 ±    0.1043       0.02868 ±    0.00310       0.09292 ±    0.00161     8.642 ±  0.138 %    87.142 ±  0.205 %
+  27       6.9623 ±    0.1055       0.02829 ±    0.00303       0.09202 ±    0.00157     8.538 ±  0.135 %    87.129 ±  0.201 %
+  28       7.0526 ±    0.1053       0.02832 ±    0.00296       0.09127 ±    0.00153     8.496 ±  0.133 %    87.149 ±  0.198 %
+  29       7.0500 ±    0.1034       0.02880 ±    0.00292       0.09137 ±    0.00149     8.557 ±  0.131 %    87.191 ±  0.194 %
+  30       7.0006 ±    0.1008       0.02977 ±    0.00287       0.09152 ±    0.00145     8.567 ±  0.129 %    87.165 ±  0.191 %
+  31       6.9004 ±    0.0974       0.02987 ±    0.00281       0.09083 ±    0.00141     8.530 ±  0.126 %    87.251 ±  0.187 %
+  32       6.7986 ±    0.0943       0.02926 ±    0.00280       0.09330 ±    0.00152     8.688 ±  0.125 %    87.225 ±  0.184 %
+  33       6.7336 ±    0.0917       0.02953 ±    0.00275       0.09274 ±    0.00149     8.677 ±  0.123 %    87.266 ±  0.181 %
+  34       6.7179 ±    0.0899       0.02937 ±    0.00268       0.09177 ±    0.00145     8.617 ±  0.121 %    87.333 ±  0.178 %
+  35       6.7270 ±    0.0888       0.02891 ±    0.00263       0.09111 ±    0.00141     8.570 ±  0.118 %    87.398 ±  0.175 %
+  36       6.7405 ±    0.0878       0.02844 ±    0.00259       0.09075 ±    0.00138     8.522 ±  0.116 %    87.414 ±  0.173 %
+  37       6.6432 ±    0.0851       0.02769 ±    0.00255       0.08997 ±    0.00135     8.495 ±  0.114 %    87.409 ±  0.171 %
+  38       6.5666 ±    0.0827       0.02662 ±    0.00252       0.09001 ±    0.00132     8.514 ±  0.112 %    87.434 ±  0.168 %
+  39       6.4867 ±    0.0803       0.02671 ±    0.00249       0.08987 ±    0.00130     8.528 ±  0.111 %    87.408 ±  0.166 %
+  40       6.3929 ±    0.0779       0.02598 ±    0.00245       0.08915 ±    0.00127     8.509 ±  0.109 %    87.424 ±  0.164 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.392929 ±   0.077851
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.97%
+Mean ln(PPL(Q)/PPL(base))     :   0.025980 ±   0.002447
+Mean PPL(Q)/PPL(base)         :   1.026321 ±   0.002511
+Mean PPL(Q)-PPL(base)         :   0.163950 ±   0.015638
+
+====== KL divergence statistics ======
+Mean    KLD:   0.089150 ±   0.001266
+Maximum KLD:  13.245127
+99.9%   KLD:   3.279078
+99.0%   KLD:   0.784911
+95.0%   KLD:   0.302743
+90.0%   KLD:   0.190118
+Median  KLD:   0.039583
+10.0%   KLD:   0.000539
+ 5.0%   KLD:   0.000127
+ 1.0%   KLD:  -0.000056
+ 0.1%   KLD:  -0.000311
+Minimum KLD:  -0.000916
+
+====== Token probability statistics ======
+Mean    Δp: -0.437 ± 0.042 %
+Maximum Δp: 97.638%
+99.9%   Δp: 52.324%
+99.0%   Δp: 22.989%
+95.0%   Δp: 10.252%
+90.0%   Δp:  5.909%
+75.0%   Δp:  1.135%
+Median  Δp: -0.001%
+25.0%   Δp: -1.497%
+10.0%   Δp: -6.935%
+ 5.0%   Δp: -12.278%
+ 1.0%   Δp: -29.036%
+ 0.1%   Δp: -71.577%
+Minimum Δp: -99.874%
+RMS Δp    :  8.509 ± 0.109 %
+Same top p: 87.424 ± 0.164 %
+

recipe/logs/N4v_kld_q106.log

@@ -0,0 +1,93 @@
+0.00.073.166 I common_init_result: fitting params to device memory ...
+0.00.073.169 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.618.000 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.632.328 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.21.360.980 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.411.524 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.21.696.824 I 
+0.21.696.976 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.21.834.422 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.25.227.075 I kl_divergence: 3.39 seconds per pass - ETA 2.25 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       6.1880 ±    0.4939       0.08329 ±    0.01978       0.13233 ±    0.00748    10.697 ±  0.585 %    86.022 ±  1.085 %
+   2       7.0357 ±    0.3926       0.05414 ±    0.01246       0.10961 ±    0.00433     9.075 ±  0.394 %    86.559 ±  0.754 %
+   3       7.4624 ±    0.3452       0.05947 ±    0.01021       0.11204 ±    0.00434     9.072 ±  0.327 %    86.250 ±  0.622 %
+   4       7.7345 ±    0.3193       0.05957 ±    0.00891       0.11467 ±    0.00464     9.213 ±  0.314 %    86.486 ±  0.535 %
+   5       7.5351 ±    0.2801       0.05356 ±    0.00785       0.11152 ±    0.00383     8.995 ±  0.270 %    86.667 ±  0.475 %
+   6       6.4595 ±    0.2123       0.04949 ±    0.00714       0.11027 ±    0.00353     9.495 ±  0.276 %    87.113 ±  0.428 %
+   7       5.9967 ±    0.1800       0.04378 ±    0.00691       0.12283 ±    0.00457    10.169 ±  0.285 %    87.250 ±  0.394 %
+   8       5.9141 ±    0.1649       0.04224 ±    0.00637       0.12112 ±    0.00408    10.053 ±  0.261 %    87.170 ±  0.370 %
+   9       6.2093 ±    0.1636       0.03840 ±    0.00600       0.12020 ±    0.00367     9.847 ±  0.240 %    86.825 ±  0.352 %
+  10       6.3404 ±    0.1596       0.04087 ±    0.00561       0.11641 ±    0.00333     9.682 ±  0.223 %    86.823 ±  0.334 %
+  11       6.4032 ±    0.1532       0.04194 ±    0.00528       0.11306 ±    0.00305     9.498 ±  0.210 %    86.795 ±  0.319 %
+  12       6.6610 ±    0.1540       0.04308 ±    0.00496       0.10959 ±    0.00281     9.281 ±  0.198 %    86.763 ±  0.306 %
+  13       6.6960 ±    0.1485       0.04143 ±    0.00472       0.10783 ±    0.00262     9.199 ±  0.188 %    86.773 ±  0.294 %
+  14       6.7584 ±    0.1443       0.04192 ±    0.00452       0.10607 ±    0.00245     9.074 ±  0.179 %    86.657 ±  0.284 %
+  15       6.8026 ±    0.1405       0.04223 ±    0.00435       0.10576 ±    0.00235     9.049 ±  0.171 %    86.660 ±  0.274 %
+  16       6.9572 ±    0.1392       0.03797 ±    0.00421       0.10524 ±    0.00223     8.990 ±  0.164 %    86.571 ±  0.267 %
+  17       7.0033 ±    0.1354       0.03762 ±    0.00406       0.10400 ±    0.00212     8.906 ±  0.157 %    86.522 ±  0.259 %
+  18       7.0947 ±    0.1334       0.03740 ±    0.00393       0.10350 ±    0.00202     8.881 ±  0.152 %    86.532 ±  0.252 %
+  19       7.0450 ±    0.1294       0.03728 ±    0.00382       0.10187 ±    0.00193     8.789 ±  0.147 %    86.639 ±  0.244 %
+  20       6.7808 ±    0.1206       0.03838 ±    0.00381       0.10695 ±    0.00193     9.167 ±  0.147 %    86.569 ±  0.238 %
+  21       6.8071 ±    0.1180       0.03975 ±    0.00374       0.10769 ±    0.00187     9.179 ±  0.142 %    86.506 ±  0.233 %
+  22       6.8328 ±    0.1159       0.04152 ±    0.00369       0.10920 ±    0.00187     9.217 ±  0.139 %    86.501 ±  0.228 %
+  23       6.8923 ±    0.1144       0.04279 ±    0.00360       0.10882 ±    0.00181     9.197 ±  0.137 %    86.498 ±  0.223 %
+  24       6.8910 ±    0.1117       0.04255 ±    0.00351       0.10896 ±    0.00176     9.191 ±  0.134 %    86.490 ±  0.218 %
+  25       6.9237 ±    0.1100       0.04257 ±    0.00344       0.10863 ±    0.00170     9.157 ±  0.130 %    86.424 ±  0.214 %
+  26       6.8938 ±    0.1073       0.04273 ±    0.00338       0.10952 ±    0.00170     9.229 ±  0.130 %    86.435 ±  0.210 %
+  27       7.0651 ±    0.1086       0.04294 ±    0.00331       0.10883 ±    0.00167     9.149 ±  0.127 %    86.452 ±  0.206 %
+  28       7.1473 ±    0.1082       0.04166 ±    0.00323       0.10759 ±    0.00162     9.083 ±  0.125 %    86.514 ±  0.202 %
+  29       7.1505 ±    0.1064       0.04295 ±    0.00318       0.10739 ±    0.00158     9.106 ±  0.123 %    86.517 ±  0.198 %
+  30       7.0929 ±    0.1035       0.04287 ±    0.00312       0.10706 ±    0.00153     9.094 ±  0.121 %    86.452 ±  0.195 %
+  31       6.9865 ±    0.1000       0.04226 ±    0.00306       0.10620 ±    0.00149     9.057 ±  0.118 %    86.573 ±  0.191 %
+  32       6.8731 ±    0.0966       0.04016 ±    0.00304       0.10897 ±    0.00167     9.204 ±  0.119 %    86.550 ±  0.189 %
+  33       6.8112 ±    0.0941       0.04100 ±    0.00300       0.10867 ±    0.00163     9.201 ±  0.117 %    86.561 ±  0.186 %
+  34       6.7925 ±    0.0922       0.04042 ±    0.00293       0.10760 ±    0.00159     9.133 ±  0.115 %    86.594 ±  0.183 %
+  35       6.8061 ±    0.0911       0.04061 ±    0.00287       0.10675 ±    0.00155     9.093 ±  0.113 %    86.655 ±  0.180 %
+  36       6.8188 ±    0.0901       0.03999 ±    0.00283       0.10633 ±    0.00151     9.056 ±  0.111 %    86.717 ±  0.177 %
+  37       6.7235 ±    0.0873       0.03971 ±    0.00278       0.10555 ±    0.00148     9.021 ±  0.109 %    86.753 ±  0.174 %
+  38       6.6468 ±    0.0849       0.03876 ±    0.00274       0.10510 ±    0.00145     9.019 ±  0.107 %    86.762 ±  0.172 %
+  39       6.5673 ±    0.0825       0.03906 ±    0.00270       0.10487 ±    0.00142     9.014 ±  0.105 %    86.773 ±  0.170 %
+  40       6.4745 ±    0.0799       0.03865 ±    0.00266       0.10441 ±    0.00139     9.030 ±  0.104 %    86.794 ±  0.167 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.474466 ±   0.079947
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.64%
+Mean ln(PPL(Q)/PPL(base))     :   0.038654 ±   0.002665
+Mean PPL(Q)/PPL(base)         :   1.039410 ±   0.002770
+Mean PPL(Q)-PPL(base)         :   0.245487 ±   0.017468
+
+====== KL divergence statistics ======
+Mean    KLD:   0.104408 ±   0.001395
+Maximum KLD:  16.437456
+99.9%   KLD:   3.401226
+99.0%   KLD:   0.899358
+95.0%   KLD:   0.354356
+90.0%   KLD:   0.226252
+Median  KLD:   0.048009
+10.0%   KLD:   0.000697
+ 5.0%   KLD:   0.000180
+ 1.0%   KLD:  -0.000044
+ 0.1%   KLD:  -0.000336
+Minimum KLD:  -0.000813
+
+====== Token probability statistics ======
+Mean    Δp: -0.202 ± 0.045 %
+Maximum Δp: 99.758%
+99.9%   Δp: 56.495%
+99.0%   Δp: 25.703%
+95.0%   Δp: 11.848%
+90.0%   Δp:  6.930%
+75.0%   Δp:  1.471%
+Median  Δp: -0.001%
+25.0%   Δp: -1.409%
+10.0%   Δp: -7.175%
+ 5.0%   Δp: -13.010%
+ 1.0%   Δp: -31.209%
+ 0.1%   Δp: -69.269%
+Minimum Δp: -98.138%
+RMS Δp    :  9.030 ± 0.104 %
+Same top p: 86.794 ± 0.167 %
+

recipe/logs/N4v_kld_q106i.log

@@ -0,0 +1,93 @@
+0.00.033.687 I common_init_result: fitting params to device memory ...
+0.00.033.690 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.365.970 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.319.923 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.19.602.693 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.19.656.222 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.19.834.989 I 
+0.19.835.081 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.19.951.235 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.22.342.997 I kl_divergence: 2.39 seconds per pass - ETA 1.58 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       5.6983 ±    0.4331       0.00085 ±    0.01708       0.10779 ±    0.00685    10.030 ±  0.635 %    86.217 ±  1.078 %
+   2       6.7307 ±    0.3660       0.00982 ±    0.01106       0.08982 ±    0.00389     8.657 ±  0.431 %    86.950 ±  0.745 %
+   3       7.1804 ±    0.3254       0.02094 ±    0.00857       0.08982 ±    0.00369     8.264 ±  0.335 %    87.260 ±  0.602 %
+   4       7.4637 ±    0.3024       0.02393 ±    0.00766       0.09566 ±    0.00581     8.559 ±  0.352 %    87.732 ±  0.513 %
+   5       7.3233 ±    0.2674       0.02505 ±    0.00682       0.09342 ±    0.00473     8.497 ±  0.302 %    87.977 ±  0.455 %
+   6       6.3148 ±    0.2036       0.02684 ±    0.00631       0.09237 ±    0.00410     8.998 ±  0.290 %    88.368 ±  0.409 %
+   7       5.8753 ±    0.1731       0.02333 ±    0.00607       0.09805 ±    0.00414     9.342 ±  0.282 %    88.479 ±  0.377 %
+   8       5.8033 ±    0.1588       0.02333 ±    0.00556       0.09655 ±    0.00369     9.256 ±  0.258 %    88.331 ±  0.355 %
+   9       6.1048 ±    0.1583       0.02143 ±    0.00523       0.09704 ±    0.00332     9.085 ±  0.238 %    87.846 ±  0.341 %
+  10       6.2216 ±    0.1541       0.02196 ±    0.00487       0.09358 ±    0.00301     8.857 ±  0.221 %    87.869 ±  0.323 %
+  11       6.2877 ±    0.1481       0.02373 ±    0.00459       0.09124 ±    0.00277     8.771 ±  0.212 %    87.914 ±  0.307 %
+  12       6.5381 ±    0.1489       0.02446 ±    0.00432       0.08825 ±    0.00255     8.571 ±  0.200 %    87.854 ±  0.295 %
+  13       6.5779 ±    0.1436       0.02363 ±    0.00412       0.08706 ±    0.00237     8.553 ±  0.189 %    87.796 ±  0.284 %
+  14       6.6420 ±    0.1396       0.02456 ±    0.00395       0.08566 ±    0.00221     8.419 ±  0.180 %    87.767 ±  0.274 %
+  15       6.6847 ±    0.1359       0.02475 ±    0.00380       0.08492 ±    0.00208     8.367 ±  0.171 %    87.683 ±  0.265 %
+  16       6.8394 ±    0.1348       0.02090 ±    0.00368       0.08419 ±    0.00198     8.280 ±  0.165 %    87.616 ±  0.257 %
+  17       6.8859 ±    0.1312       0.02070 ±    0.00355       0.08311 ±    0.00187     8.185 ±  0.158 %    87.695 ±  0.249 %
+  18       6.9788 ±    0.1294       0.02093 ±    0.00343       0.08257 ±    0.00178     8.136 ±  0.151 %    87.640 ±  0.243 %
+  19       6.9229 ±    0.1253       0.01979 ±    0.00332       0.08171 ±    0.00170     8.036 ±  0.146 %    87.647 ±  0.236 %
+  20       6.6565 ±    0.1166       0.01988 ±    0.00330       0.08616 ±    0.00169     8.356 ±  0.141 %    87.542 ±  0.231 %
+  21       6.6723 ±    0.1139       0.01975 ±    0.00323       0.08644 ±    0.00162     8.308 ±  0.137 %    87.488 ±  0.226 %
+  22       6.6896 ±    0.1116       0.02034 ±    0.00316       0.08711 ±    0.00160     8.355 ±  0.137 %    87.514 ±  0.220 %
+  23       6.7506 ±    0.1102       0.02201 ±    0.00309       0.08713 ±    0.00156     8.332 ±  0.134 %    87.479 ±  0.216 %
+  24       6.7515 ±    0.1077       0.02210 ±    0.00303       0.08701 ±    0.00153     8.313 ±  0.132 %    87.553 ±  0.211 %
+  25       6.7898 ±    0.1062       0.02305 ±    0.00295       0.08650 ±    0.00148     8.258 ±  0.129 %    87.464 ±  0.207 %
+  26       6.7594 ±    0.1035       0.02304 ±    0.00290       0.08694 ±    0.00146     8.287 ±  0.127 %    87.540 ±  0.203 %
+  27       6.9231 ±    0.1046       0.02264 ±    0.00283       0.08601 ±    0.00141     8.197 ±  0.124 %    87.553 ±  0.199 %
+  28       7.0130 ±    0.1044       0.02268 ±    0.00277       0.08539 ±    0.00138     8.177 ±  0.123 %    87.586 ±  0.195 %
+  29       7.0144 ±    0.1026       0.02374 ±    0.00273       0.08551 ±    0.00135     8.241 ±  0.122 %    87.569 ±  0.192 %
+  30       6.9699 ±    0.1001       0.02537 ±    0.00270       0.08585 ±    0.00132     8.254 ±  0.120 %    87.563 ±  0.188 %
+  31       6.8694 ±    0.0967       0.02536 ±    0.00264       0.08520 ±    0.00128     8.219 ±  0.117 %    87.668 ±  0.185 %
+  32       6.7608 ±    0.0936       0.02369 ±    0.00262       0.08761 ±    0.00146     8.350 ±  0.118 %    87.662 ±  0.182 %
+  33       6.6975 ±    0.0910       0.02416 ±    0.00258       0.08714 ±    0.00143     8.347 ±  0.117 %    87.704 ±  0.179 %
+  34       6.6852 ±    0.0894       0.02449 ±    0.00252       0.08623 ±    0.00139     8.290 ±  0.114 %    87.775 ±  0.176 %
+  35       6.6941 ±    0.0882       0.02401 ±    0.00247       0.08558 ±    0.00135     8.241 ±  0.112 %    87.845 ±  0.173 %
+  36       6.7094 ±    0.0873       0.02381 ±    0.00243       0.08528 ±    0.00132     8.207 ±  0.110 %    87.846 ±  0.170 %
+  37       6.6134 ±    0.0846       0.02319 ±    0.00240       0.08448 ±    0.00129     8.182 ±  0.108 %    87.852 ±  0.168 %
+  38       6.5369 ±    0.0822       0.02208 ±    0.00237       0.08434 ±    0.00126     8.180 ±  0.106 %    87.897 ±  0.165 %
+  39       6.4545 ±    0.0798       0.02173 ±    0.00234       0.08418 ±    0.00123     8.209 ±  0.105 %    87.896 ±  0.163 %
+  40       6.3642 ±    0.0773       0.02147 ±    0.00231       0.08356 ±    0.00121     8.201 ±  0.103 %    87.903 ±  0.161 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.364168 ±   0.077334
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  98.19%
+Mean ln(PPL(Q)/PPL(base))     :   0.021471 ±   0.002306
+Mean PPL(Q)/PPL(base)         :   1.021703 ±   0.002356
+Mean PPL(Q)-PPL(base)         :   0.135189 ±   0.014650
+
+====== KL divergence statistics ======
+Mean    KLD:   0.083563 ±   0.001206
+Maximum KLD:  15.989422
+99.9%   KLD:   2.686481
+99.0%   KLD:   0.713036
+95.0%   KLD:   0.286778
+90.0%   KLD:   0.180925
+Median  KLD:   0.038000
+10.0%   KLD:   0.000531
+ 5.0%   KLD:   0.000125
+ 1.0%   KLD:  -0.000041
+ 0.1%   KLD:  -0.000270
+Minimum KLD:  -0.000789
+
+====== Token probability statistics ======
+Mean    Δp: -0.435 ± 0.040 %
+Maximum Δp: 99.469%
+99.9%   Δp: 52.136%
+99.0%   Δp: 22.860%
+95.0%   Δp:  9.882%
+90.0%   Δp:  5.626%
+75.0%   Δp:  1.034%
+Median  Δp: -0.004%
+25.0%   Δp: -1.525%
+10.0%   Δp: -6.894%
+ 5.0%   Δp: -11.769%
+ 1.0%   Δp: -27.478%
+ 0.1%   Δp: -66.247%
+Minimum Δp: -99.814%
+RMS Δp    :  8.201 ± 0.103 %
+Same top p: 87.903 ± 0.161 %
+

recipe/logs/N5_kld_q106_repeat.log

@@ -0,0 +1,92 @@
+0.00.038.287 I common_init_result: fitting params to device memory ...
+0.00.038.289 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.399.780 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.682.112 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.729.253 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.921.065 I 
+0.20.921.189 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.21.035.122 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.22.991.105 I kl_divergence: 1.96 seconds per pass - ETA 1.30 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       6.2617 ±    0.5005       0.09513 ±    0.01958       0.13331 ±    0.00723    11.273 ±  0.670 %    85.728 ±  1.094 %
+   2       7.1013 ±    0.3976       0.06343 ±    0.01246       0.11005 ±    0.00427     9.422 ±  0.440 %    86.413 ±  0.758 %
+   3       7.5058 ±    0.3468       0.06527 ±    0.01057       0.11313 ±    0.00423     9.250 ±  0.346 %    86.022 ±  0.626 %
+   4       7.7849 ±    0.3204       0.06605 ±    0.00908       0.11249 ±    0.00383     9.264 ±  0.306 %    86.388 ±  0.536 %
+   5       7.5801 ±    0.2810       0.05952 ±    0.00798       0.11043 ±    0.00322     9.087 ±  0.266 %    86.725 ±  0.474 %
+   6       6.5014 ±    0.2131       0.05596 ±    0.00724       0.11026 ±    0.00320     9.619 ±  0.279 %    87.178 ±  0.427 %
+   7       6.0313 ±    0.1806       0.04953 ±    0.00696       0.12172 ±    0.00436    10.069 ±  0.275 %    87.236 ±  0.394 %
+   8       5.9472 ±    0.1655       0.04783 ±    0.00647       0.12058 ±    0.00391     9.949 ±  0.252 %    87.121 ±  0.370 %
+   9       6.2446 ±    0.1643       0.04408 ±    0.00605       0.12012 ±    0.00353     9.791 ±  0.233 %    86.597 ±  0.355 %
+  10       6.3764 ±    0.1603       0.04654 ±    0.00565       0.11624 ±    0.00321     9.609 ±  0.217 %    86.588 ±  0.337 %
+  11       6.4374 ±    0.1538       0.04726 ±    0.00530       0.11268 ±    0.00294     9.442 ±  0.204 %    86.484 ±  0.322 %
+  12       6.6950 ±    0.1547       0.04816 ±    0.00499       0.10924 ±    0.00271     9.238 ±  0.193 %    86.388 ±  0.310 %
+  13       6.7262 ±    0.1490       0.04592 ±    0.00474       0.10773 ±    0.00253     9.174 ±  0.183 %    86.413 ±  0.297 %
+  14       6.7857 ±    0.1447       0.04595 ±    0.00454       0.10583 ±    0.00237     9.050 ±  0.174 %    86.426 ±  0.286 %
+  15       6.8279 ±    0.1408       0.04594 ±    0.00437       0.10522 ±    0.00224     9.021 ±  0.166 %    86.419 ±  0.277 %
+  16       6.9800 ±    0.1394       0.04125 ±    0.00424       0.10488 ±    0.00213     8.960 ±  0.160 %    86.339 ±  0.268 %
+  17       7.0246 ±    0.1356       0.04064 ±    0.00408       0.10348 ±    0.00202     8.863 ±  0.153 %    86.424 ±  0.260 %
+  18       7.1128 ±    0.1335       0.03994 ±    0.00395       0.10297 ±    0.00192     8.819 ±  0.147 %    86.385 ±  0.253 %
+  19       7.0597 ±    0.1294       0.03935 ±    0.00383       0.10135 ±    0.00184     8.726 ±  0.143 %    86.474 ±  0.245 %
+  20       6.7990 ±    0.1207       0.04105 ±    0.00383       0.10693 ±    0.00185     9.139 ±  0.143 %    86.393 ±  0.240 %
+  21       6.8267 ±    0.1182       0.04262 ±    0.00375       0.10765 ±    0.00179     9.144 ±  0.139 %    86.375 ±  0.234 %
+  22       6.8456 ±    0.1159       0.04339 ±    0.00369       0.10893 ±    0.00181     9.184 ±  0.137 %    86.417 ±  0.228 %
+  23       6.9000 ±    0.1143       0.04390 ±    0.00360       0.10839 ±    0.00175     9.147 ±  0.133 %    86.383 ±  0.224 %
+  24       6.8942 ±    0.1115       0.04302 ±    0.00351       0.10851 ±    0.00171     9.165 ±  0.132 %    86.376 ±  0.219 %
+  25       6.9297 ±    0.1099       0.04345 ±    0.00344       0.10819 ±    0.00165     9.128 ±  0.128 %    86.334 ±  0.215 %
+  26       6.8989 ±    0.1072       0.04347 ±    0.00338       0.10899 ±    0.00166     9.185 ±  0.127 %    86.371 ±  0.210 %
+  27       7.0668 ±    0.1084       0.04318 ±    0.00330       0.10805 ±    0.00160     9.103 ±  0.124 %    86.384 ±  0.206 %
+  28       7.1505 ±    0.1080       0.04210 ±    0.00321       0.10685 ±    0.00155     9.028 ±  0.122 %    86.395 ±  0.203 %
+  29       7.1560 ±    0.1063       0.04372 ±    0.00317       0.10671 ±    0.00152     9.048 ±  0.120 %    86.392 ±  0.199 %
+  30       7.0985 ±    0.1034       0.04365 ±    0.00312       0.10649 ±    0.00148     9.044 ±  0.118 %    86.341 ±  0.196 %
+  31       6.9900 ±    0.0998       0.04276 ±    0.00305       0.10563 ±    0.00144     9.018 ±  0.116 %    86.444 ±  0.192 %
+  32       6.8747 ±    0.0965       0.04039 ±    0.00305       0.10917 ±    0.00170     9.201 ±  0.119 %    86.416 ±  0.189 %
+  33       6.8130 ±    0.0939       0.04125 ±    0.00301       0.10876 ±    0.00166     9.199 ±  0.117 %    86.457 ±  0.186 %
+  34       6.7927 ±    0.0920       0.04044 ±    0.00294       0.10768 ±    0.00161     9.128 ±  0.115 %    86.441 ±  0.184 %
+  35       6.8056 ±    0.0909       0.04054 ±    0.00288       0.10692 ±    0.00157     9.083 ±  0.113 %    86.494 ±  0.181 %
+  36       6.8222 ±    0.0899       0.04049 ±    0.00284       0.10659 ±    0.00154     9.060 ±  0.111 %    86.546 ±  0.178 %
+  37       6.7263 ±    0.0872       0.04013 ±    0.00279       0.10571 ±    0.00150     9.033 ±  0.109 %    86.566 ±  0.175 %
+  38       6.6472 ±    0.0847       0.03882 ±    0.00275       0.10525 ±    0.00147     9.020 ±  0.107 %    86.621 ±  0.173 %
+  39       6.5675 ±    0.0823       0.03909 ±    0.00271       0.10501 ±    0.00144     9.022 ±  0.106 %    86.638 ±  0.170 %
+  40       6.4740 ±    0.0798       0.03858 ±    0.00267       0.10444 ±    0.00141     9.028 ±  0.105 %    86.659 ±  0.168 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.473964 ±   0.079799
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.63%
+Mean ln(PPL(Q)/PPL(base))     :   0.038576 ±   0.002667
+Mean PPL(Q)/PPL(base)         :   1.039330 ±   0.002772
+Mean PPL(Q)-PPL(base)         :   0.244985 ±   0.017453
+
+====== KL divergence statistics ======
+Mean    KLD:   0.104436 ±   0.001413
+Maximum KLD:  13.909879
+99.9%   KLD:   3.189192
+99.0%   KLD:   0.926687
+95.0%   KLD:   0.354656
+90.0%   KLD:   0.225739
+Median  KLD:   0.047702
+10.0%   KLD:   0.000698
+ 5.0%   KLD:   0.000183
+ 1.0%   KLD:  -0.000033
+ 0.1%   KLD:  -0.000321
+Minimum KLD:  -0.000718
+
+====== Token probability statistics ======
+Mean    Δp: -0.244 ± 0.045 %
+Maximum Δp: 99.706%
+99.9%   Δp: 59.493%
+99.0%   Δp: 25.100%
+95.0%   Δp: 11.707%
+90.0%   Δp:  6.963%
+75.0%   Δp:  1.453%
+Median  Δp: -0.001%
+25.0%   Δp: -1.479%
+10.0%   Δp: -7.240%
+ 5.0%   Δp: -12.992%
+ 1.0%   Δp: -31.268%
+ 0.1%   Δp: -68.467%
+Minimum Δp: -98.412%
+RMS Δp    :  9.028 ± 0.105 %
+Same top p: 86.659 ± 0.168 %
+

recipe/logs/N5v_kld_q106_repeat.log

@@ -0,0 +1,93 @@
+0.00.034.429 I common_init_result: fitting params to device memory ...
+0.00.034.432 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.372.660 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.01.317.954 W read_raw_unsafe: Falling back to buffered IO due to Bad address
+0.19.648.031 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.19.696.562 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.19.866.239 I 
+0.19.866.320 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.19.981.669 I kl_divergence: computing over 40 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.22.378.306 I kl_divergence: 2.40 seconds per pass - ETA 1.58 minutes
+
+chunk             PPL               ln(PPL(Q)/PPL(base))          KL Divergence              Δp RMS            Same top p
+   1       6.1880 ±    0.4939       0.08329 ±    0.01978       0.13233 ±    0.00748    10.697 ±  0.585 %    86.022 ±  1.085 %
+   2       7.0357 ±    0.3926       0.05414 ±    0.01246       0.10961 ±    0.00433     9.075 ±  0.394 %    86.559 ±  0.754 %
+   3       7.4624 ±    0.3452       0.05947 ±    0.01021       0.11204 ±    0.00434     9.072 ±  0.327 %    86.250 ±  0.622 %
+   4       7.7345 ±    0.3193       0.05957 ±    0.00891       0.11467 ±    0.00464     9.213 ±  0.314 %    86.486 ±  0.535 %
+   5       7.5351 ±    0.2801       0.05356 ±    0.00785       0.11152 ±    0.00383     8.995 ±  0.270 %    86.667 ±  0.475 %
+   6       6.4595 ±    0.2123       0.04949 ±    0.00714       0.11027 ±    0.00353     9.495 ±  0.276 %    87.113 ±  0.428 %
+   7       5.9967 ±    0.1800       0.04378 ±    0.00691       0.12283 ±    0.00457    10.169 ±  0.285 %    87.250 ±  0.394 %
+   8       5.9141 ±    0.1649       0.04224 ±    0.00637       0.12112 ±    0.00408    10.053 ±  0.261 %    87.170 ±  0.370 %
+   9       6.2093 ±    0.1636       0.03840 ±    0.00600       0.12020 ±    0.00367     9.847 ±  0.240 %    86.825 ±  0.352 %
+  10       6.3404 ±    0.1596       0.04087 ±    0.00561       0.11641 ±    0.00333     9.682 ±  0.223 %    86.823 ±  0.334 %
+  11       6.4032 ±    0.1532       0.04194 ±    0.00528       0.11306 ±    0.00305     9.498 ±  0.210 %    86.795 ±  0.319 %
+  12       6.6610 ±    0.1540       0.04308 ±    0.00496       0.10959 ±    0.00281     9.281 ±  0.198 %    86.763 ±  0.306 %
+  13       6.6960 ±    0.1485       0.04143 ±    0.00472       0.10783 ±    0.00262     9.199 ±  0.188 %    86.773 ±  0.294 %
+  14       6.7584 ±    0.1443       0.04192 ±    0.00452       0.10607 ±    0.00245     9.074 ±  0.179 %    86.657 ±  0.284 %
+  15       6.8026 ±    0.1405       0.04223 ±    0.00435       0.10576 ±    0.00235     9.049 ±  0.171 %    86.660 ±  0.274 %
+  16       6.9572 ±    0.1392       0.03797 ±    0.00421       0.10524 ±    0.00223     8.990 ±  0.164 %    86.571 ±  0.267 %
+  17       7.0033 ±    0.1354       0.03762 ±    0.00406       0.10400 ±    0.00212     8.906 ±  0.157 %    86.522 ±  0.259 %
+  18       7.0947 ±    0.1334       0.03740 ±    0.00393       0.10350 ±    0.00202     8.881 ±  0.152 %    86.532 ±  0.252 %
+  19       7.0450 ±    0.1294       0.03728 ±    0.00382       0.10187 ±    0.00193     8.789 ±  0.147 %    86.639 ±  0.244 %
+  20       6.7808 ±    0.1206       0.03838 ±    0.00381       0.10695 ±    0.00193     9.167 ±  0.147 %    86.569 ±  0.238 %
+  21       6.8071 ±    0.1180       0.03975 ±    0.00374       0.10769 ±    0.00187     9.179 ±  0.142 %    86.506 ±  0.233 %
+  22       6.8328 ±    0.1159       0.04152 ±    0.00369       0.10920 ±    0.00187     9.217 ±  0.139 %    86.501 ±  0.228 %
+  23       6.8923 ±    0.1144       0.04279 ±    0.00360       0.10882 ±    0.00181     9.197 ±  0.137 %    86.498 ±  0.223 %
+  24       6.8910 ±    0.1117       0.04255 ±    0.00351       0.10896 ±    0.00176     9.191 ±  0.134 %    86.490 ±  0.218 %
+  25       6.9237 ±    0.1100       0.04257 ±    0.00344       0.10863 ±    0.00170     9.157 ±  0.130 %    86.424 ±  0.214 %
+  26       6.8938 ±    0.1073       0.04273 ±    0.00338       0.10952 ±    0.00170     9.229 ±  0.130 %    86.435 ±  0.210 %
+  27       7.0651 ±    0.1086       0.04294 ±    0.00331       0.10883 ±    0.00167     9.149 ±  0.127 %    86.452 ±  0.206 %
+  28       7.1473 ±    0.1082       0.04166 ±    0.00323       0.10759 ±    0.00162     9.083 ±  0.125 %    86.514 ±  0.202 %
+  29       7.1505 ±    0.1064       0.04295 ±    0.00318       0.10739 ±    0.00158     9.106 ±  0.123 %    86.517 ±  0.198 %
+  30       7.0929 ±    0.1035       0.04287 ±    0.00312       0.10706 ±    0.00153     9.094 ±  0.121 %    86.452 ±  0.195 %
+  31       6.9865 ±    0.1000       0.04226 ±    0.00306       0.10620 ±    0.00149     9.057 ±  0.118 %    86.573 ±  0.191 %
+  32       6.8731 ±    0.0966       0.04016 ±    0.00304       0.10897 ±    0.00167     9.204 ±  0.119 %    86.550 ±  0.189 %
+  33       6.8112 ±    0.0941       0.04100 ±    0.00300       0.10867 ±    0.00163     9.201 ±  0.117 %    86.561 ±  0.186 %
+  34       6.7925 ±    0.0922       0.04042 ±    0.00293       0.10760 ±    0.00159     9.133 ±  0.115 %    86.594 ±  0.183 %
+  35       6.8061 ±    0.0911       0.04061 ±    0.00287       0.10675 ±    0.00155     9.093 ±  0.113 %    86.655 ±  0.180 %
+  36       6.8188 ±    0.0901       0.03999 ±    0.00283       0.10633 ±    0.00151     9.056 ±  0.111 %    86.717 ±  0.177 %
+  37       6.7235 ±    0.0873       0.03971 ±    0.00278       0.10555 ±    0.00148     9.021 ±  0.109 %    86.753 ±  0.174 %
+  38       6.6468 ±    0.0849       0.03876 ±    0.00274       0.10510 ±    0.00145     9.019 ±  0.107 %    86.762 ±  0.172 %
+  39       6.5673 ±    0.0825       0.03906 ±    0.00270       0.10487 ±    0.00142     9.014 ±  0.105 %    86.773 ±  0.170 %
+  40       6.4745 ±    0.0799       0.03865 ±    0.00266       0.10441 ±    0.00139     9.030 ±  0.104 %    86.794 ±  0.167 %
+
+====== Perplexity statistics ======
+Mean PPL(Q)                   :   6.474466 ±   0.079947
+Mean PPL(base)                :   6.228979 ±   0.075322
+Cor(ln(PPL(Q)), ln(PPL(base))):  97.64%
+Mean ln(PPL(Q)/PPL(base))     :   0.038654 ±   0.002665
+Mean PPL(Q)/PPL(base)         :   1.039410 ±   0.002770
+Mean PPL(Q)-PPL(base)         :   0.245487 ±   0.017468
+
+====== KL divergence statistics ======
+Mean    KLD:   0.104408 ±   0.001395
+Maximum KLD:  16.437456
+99.9%   KLD:   3.401226
+99.0%   KLD:   0.899358
+95.0%   KLD:   0.354356
+90.0%   KLD:   0.226252
+Median  KLD:   0.048009
+10.0%   KLD:   0.000697
+ 5.0%   KLD:   0.000180
+ 1.0%   KLD:  -0.000044
+ 0.1%   KLD:  -0.000336
+Minimum KLD:  -0.000813
+
+====== Token probability statistics ======
+Mean    Δp: -0.202 ± 0.045 %
+Maximum Δp: 99.758%
+99.9%   Δp: 56.495%
+99.0%   Δp: 25.703%
+95.0%   Δp: 11.848%
+90.0%   Δp:  6.930%
+75.0%   Δp:  1.471%
+Median  Δp: -0.001%
+25.0%   Δp: -1.409%
+10.0%   Δp: -7.175%
+ 5.0%   Δp: -13.010%
+ 1.0%   Δp: -31.209%
+ 0.1%   Δp: -69.269%
+Minimum Δp: -98.138%
+RMS Δp    :  9.030 ± 0.104 %
+Same top p: 86.794 ± 0.167 %
+

recipe/logs/N6t_tools_c1.log

@@ -0,0 +1,33 @@
+  PASS think=True  multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  FAIL think=True  nested-object: exception KeyError('tool_calls')
+  PASS think=True  enum: unit=fahrenheit
+  PASS think=True  correct-decline: content='391'
+  PASS think=True  multi-turn: final='Tokyo is **21°C** with **clear skies**.'
+  PASS think=True  streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  PASS think=True  parallel: calls=['lima', 'oslo']
+  PASS think=False multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  PASS think=False nested-object: args={'title': 'Design review', 'when': {'date': '2026-10-02', 'time': '14:00'}, 'attendees': ['ana@x.io', 'bo@x.io']}
+  PASS think=False enum: unit=fahrenheit
+  PASS think=False correct-decline: content='391'
+  PASS think=False multi-turn: final='Tokyo is currently **21°C** with **clear skies**.'
+  PASS think=False streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  PASS think=False parallel: calls=['lima', 'oslo']
+{"label": "n-tools-q106-c1", "passed": 13, "total": 14, "detail": {"multi-arg|think=True": true, "nested-object|think=True": false, "enum|think=True": true, "correct-decline|think=True": true, "multi-turn|think=True": true, "streaming|think=True": true, "parallel|think=True": true, "multi-arg|think=False": true, "nested-object|think=False": true, "enum|think=False": true, "correct-decline|think=False": true, "multi-turn|think=False": true, "streaming|think=False": true, "parallel|think=False": true}}
+{"label": "n-vision-q106-c1-faon", "fa": "on", "mtp": false, "expected": "red,blue,circle,square", "answer": "The image shows two shapes: a red circle on the left and a blue square on the right.", "hits": ["red", "blue", "circle", "square"], "error": null, "server_died": false, "server_log_errors": [], "result": "PASS"}
+probe no-kwargs correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe no-kwargs single-word {"content": "ready", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe no-kwargs multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+probe enable_thinking=false correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe enable_thinking=false single-word {"content": "ready", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe enable_thinking=false multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+probe reasoning_effort=high correct-decline {"content": "We need answer directly. 391.\n</think>\n\n391", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=high single-word {"content": "We need need output exactly ready.\n</think>\n\nready", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=high multi-arg {"content": "We need need tool. Current weather Paris celsius.\n</think>\n\n", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": ["</think>"]}
+probe reasoning_effort=medium correct-decline {"content": "\n\n</think>\n\n391", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=medium single-word {"content": "\n\n</think>\n\nready", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=medium multi-arg {"content": "\n\n</think>\n\n", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": ["</think>"]}
+probe reasoning_effort=none correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe reasoning_effort=none single-word {"content": "ready", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe reasoning_effort=none multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+NEX_TOOLS_TPL_DONE
+rc=0

recipe/logs/N6t_tools_tpl.log

@@ -0,0 +1,22 @@
+  PASS think=True  multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  PASS think=True  nested-object: args={'title': 'Design review', 'when': {'date': '2026-10-02', 'time': '14:00'}, 'attendees': ['ana@x.io', 'bo@x.io']}
+  PASS think=True  enum: unit=fahrenheit
+  PASS think=True  correct-decline: content='391'
+  PASS think=True  multi-turn: final='Tokyo is currently **21°C and clear**.'
+  PASS think=True  streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  PASS think=True  parallel: calls=['lima', 'oslo']
+  PASS think=False multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  PASS think=False nested-object: args={'title': 'Design review', 'when': {'date': '2026-10-02', 'time': '14:00'}, 'attendees': ['ana@x.io', 'bo@x.io']}
+  PASS think=False enum: unit=fahrenheit
+  PASS think=False correct-decline: content='391'
+  PASS think=False multi-turn: final='Tokyo is currently **21°C**, with **clear skies**.'
+  PASS think=False streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  PASS think=False parallel: calls=['lima', 'oslo']
+{"label": "n-tools-q106-tpl", "passed": 14, "total": 14, "detail": {"multi-arg|think=True": true, "nested-object|think=True": true, "enum|think=True": true, "correct-decline|think=True": true, "multi-turn|think=True": true, "streaming|think=True": true, "parallel|think=True": true, "multi-arg|think=False": true, "nested-object|think=False": true, "enum|think=False": true, "correct-decline|think=False": true, "multi-turn|think=False": true, "streaming|think=False": true, "parallel|think=False": true}}
+probe no-kwargs correct-decline {"content": "391", "reasoning_len": 30, "tool_calls": [], "leaks": []}
+probe no-kwargs multi-arg {"content": "", "reasoning_len": 50, "tool_calls": ["get_weather"], "leaks": []}
+probe reasoning_effort=medium correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe reasoning_effort=medium multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+probe reasoning_effort=none correct-decline {"content": "", "reasoning_len": 3, "tool_calls": [], "leaks": []}
+probe reasoning_effort=none multi-arg {"content": "", "reasoning_len": 133, "tool_calls": [], "leaks": []}
+NEX_TOOLS_TPL_DONE

recipe/logs/N6t_tools_tpl_medium.log

@@ -0,0 +1,31 @@
+  FAIL think=True  multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  FAIL think=True  nested-object: exception KeyError('tool_calls')
+  FAIL think=True  enum: unit=fahrenheit
+  FAIL think=True  correct-decline: content='</think>\n\n391'
+  FAIL think=True  multi-turn: final='</think>\n\nTokyo is currently **21°C** and **clear**.'
+  FAIL think=True  streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  FAIL think=True  parallel: calls=['lima', 'oslo']
+  PASS think=False multi-arg: args={'city': 'Paris', 'unit': 'celsius'}
+  PASS think=False nested-object: args={'title': 'Design review', 'when': {'date': '2026-10-02', 'time': '14:00'}, 'attendees': ['ana@x.io', 'bo@x.io']}
+  PASS think=False enum: unit=fahrenheit
+  PASS think=False correct-decline: content='391'
+  PASS think=False multi-turn: final='Tokyo is currently **21°C** with **clear skies**.'
+  PASS think=False streaming: stream args={'city': 'Rome', 'unit': 'celsius'}
+  PASS think=False parallel: calls=['lima', 'oslo']
+{"label": "n-tools-q106-tpl-medium", "passed": 7, "total": 14, "detail": {"multi-arg|think=True": false, "nested-object|think=True": false, "enum|think=True": false, "correct-decline|think=True": false, "multi-turn|think=True": false, "streaming|think=True": false, "parallel|think=True": false, "multi-arg|think=False": true, "nested-object|think=False": true, "enum|think=False": true, "correct-decline|think=False": true, "multi-turn|think=False": true, "streaming|think=False": true, "parallel|think=False": true}}
+probe no-kwargs correct-decline {"content": "</think>\n\n391", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe no-kwargs single-word {"content": "</think>\n\nready", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe no-kwargs multi-arg {"content": "</think>\n\n", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": ["</think>"]}
+probe enable_thinking=false correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe enable_thinking=false single-word {"content": "ready", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe enable_thinking=false multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+probe reasoning_effort=high correct-decline {"content": "We need answer directly. 391.\n</think>\n\n391", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=high single-word {"content": "We need need output exactly ready.\n</think>\n\nready", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=high multi-arg {"content": "We need need tool. Current weather Paris celsius.\n</think>\n\n", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": ["</think>"]}
+probe reasoning_effort=medium correct-decline {"content": "</think>\n\n391", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=medium single-word {"content": "</think>\n\nready", "reasoning_len": 0, "tool_calls": [], "leaks": ["</think>"]}
+probe reasoning_effort=medium multi-arg {"content": "</think>\n\n", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": ["</think>"]}
+probe reasoning_effort=none correct-decline {"content": "391", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe reasoning_effort=none single-word {"content": "ready", "reasoning_len": 0, "tool_calls": [], "leaks": []}
+probe reasoning_effort=none multi-arg {"content": "", "reasoning_len": 0, "tool_calls": ["get_weather"], "leaks": []}
+NEX_TOOLS_TPL_DONE

recipe/logs/N7_sizing.log

@@ -0,0 +1,5 @@
+[2026-09-17T00:50:08Z] waiting for the quiet-box lock
+[2026-09-17T00:50:08Z] quiet-box lock held
+{"label":"strix-lean","ctx":65536,"avail_before":122.34,"footprint_loaded_gib":21.11,"footprint_after_8k_gib":21.29}
+{"label":"strix-lean","ctx":262144,"avail_before":122.15,"footprint_loaded_gib":24.36,"footprint_after_8k_gib":24.52}
+NEX_SIZING_DONE

recipe/logs/N8_unice.log

@@ -0,0 +1,3 @@
+[2026-09-17T00:50:56Z] Nex FAST test seats: create + smoke test (box still iced)
+[2026-09-17T00:51:38Z]   nex_seats exit=0 ([2026-09-17T00:51:38Z] NEX_SEATS_DONE fail=0)
+[2026-09-17T00:51:38Z] UNICE_DEFERRED: OxCoder-9B build (King 2026-09-16 22:33Z: "when that is done do this one"); its final step un-ices

recipe/logs/N8a_seats.log

@@ -0,0 +1,8 @@
+[2026-09-17T00:50:56Z] plan: [('max1-nex-fast', 'ROCm0', 'fa on', 262144, '31000M', True), ('max1-nex-fast-imat', 'ROCm0', 'fa on', 262144, '31000M', True)]
+[2026-09-17T00:50:56Z] max1-nex-fast written (ROCm0, -fa on, ctx 262144, MemoryMax 31000M) -> smoke test
+{"unit": "max1-nex-fast", "port": 8097, "load_s": 25, "time": "2026-09-17T00:51:21Z", "direct_reply": "ready", "direct_tg": 92.93680297397769, "gateway_model": "nex-n2.5-mini-fast@max1", "gateway_reply": "ready", "result": "PASS"}
+[2026-09-17T00:51:22Z]   max1-nex-fast stopped (enabled: disabled)
+[2026-09-17T00:51:27Z] max1-nex-fast-imat written (ROCm0, -fa on, ctx 262144, MemoryMax 31000M) -> smoke test
+{"unit": "max1-nex-fast-imat", "port": 8098, "load_s": 5, "time": "2026-09-17T00:51:32Z", "direct_reply": "ready", "direct_tg": 94.37078280564337, "gateway_model": "nex-n2.5-mini-fast-imatrix@max1", "gateway_reply": "ready", "result": "PASS"}
+[2026-09-17T00:51:33Z]   max1-nex-fast-imat stopped (enabled: disabled)
+[2026-09-17T00:51:38Z] NEX_SEATS_DONE fail=0

recipe/logs/N8d_seats.log

@@ -0,0 +1,8 @@
+[2026-09-17T01:20:44Z] plan: [('max1-nex-fast', 'ROCm0', 'fa on', 262144, '31000M', True), ('max1-nex-fast-imat', 'ROCm0', 'fa on', 262144, '31000M', True)]
+[2026-09-17T01:20:45Z] max1-nex-fast written (ROCm0, -fa on, ctx 262144, MemoryMax 31000M) -> smoke test
+{"unit": "max1-nex-fast", "port": 8097, "load_s": 25, "time": "2026-09-17T01:21:10Z", "direct_reply": "ready", "direct_tg": 41.41386950489719, "default_reply": "ready", "default_reasoning_len": 0, "default_leak": false, "thinking_reply": "", "thinking_reasoning_len": 5, "thinking_leak": false, "gateway_model": "nex-n2.5-mini-fast@max1", "gateway_reply": "ready", "result": "PASS"}
+[2026-09-17T01:21:12Z]   max1-nex-fast stopped (enabled: disabled)
+[2026-09-17T01:21:17Z] max1-nex-fast-imat written (ROCm0, -fa on, ctx 262144, MemoryMax 31000M) -> smoke test
+{"unit": "max1-nex-fast-imat", "port": 8098, "load_s": 25, "time": "2026-09-17T01:21:43Z", "direct_reply": "ready", "direct_tg": 41.54290343352097, "default_reply": "ready", "default_reasoning_len": 0, "default_leak": false, "thinking_reply": "", "thinking_reasoning_len": 5, "thinking_leak": false, "gateway_model": "nex-n2.5-mini-fast-imatrix@max1", "gateway_reply": "ready", "result": "PASS"}
+[2026-09-17T01:21:44Z]   max1-nex-fast-imat stopped (enabled: disabled)
+[2026-09-17T01:21:49Z] NEX_SEATS_DONE fail=0

recipe/logs/Q_readback.log

@@ -0,0 +1,3 @@
+PASS Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf arch=qwen35moe ftype=106 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q5_K
+PASS Nex-N2.5-mini-Q4_0_ROCMFP4_COHERENT.gguf arch=qwen35moe ftype=102 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q6_K
+PASS Nex-N2.5-mini-Q4_0_ROCMFP4_FAST.gguf arch=qwen35moe ftype=103 tensors=733 nextn=0 output.weight=Q6_K token_embd.weight=Q4_0_ROCMFP4_FAST

recipe/logs/Q_sizes.log

@@ -0,0 +1,6 @@
+Q1_q106  quant size  = 17865.52 MiB (4.32 BPW)
+Q1_q102  quant size  = 18916.30 MiB (4.58 BPW)
+Q1_q103  quant size  = 17774.11 MiB (4.30 BPW)
+N3_q106i  quant size  = 17865.52 MiB (4.32 BPW)
+N3_q102i  quant size  = 18916.30 MiB (4.58 BPW)
+N3_q103i  quant size  = 17774.11 MiB (4.30 BPW)

recipe/logs/b_n-c3-q106.log

@@ -0,0 +1,284 @@
+0.00.041.768 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.041.771 I device_info:
+0.00.041.819 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 125259 MiB free)
+0.00.041.885 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131926 MiB free)
+0.00.041.888 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.041.930 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.041.949 I srv          init: running without SSL
+0.00.041.972 I srv          init: using 31 threads for HTTP server
+0.00.041.973 I srv          init: the WebUI is disabled
+0.00.042.029 I srv         start: binding port with default address family
+0.00.043.171 I srv          main: loading model
+0.00.043.173 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.076.857 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.443.494 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.605.862 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.804.349 I srv    load_model: initializing slots, n_slots = 1
+0.20.970.363 W srv    load_model: speculative decoding will use checkpoints
+0.20.970.368 W common_speculative_init: no implementations specified for speculative decoding
+0.20.970.369 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.20.970.402 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.20.970.414 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.20.970.429 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.20.979.135 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.20.985.480 I srv          init: init: chat template, thinking = 1
+0.20.985.501 I srv          main: model loaded
+0.20.985.504 I srv          main: server is listening on http://127.0.0.1:18600
+0.20.985.506 I srv  update_slots: all slots are idle
+0.23.012.828 I srv  params_from_: Chat format: peg-native
+0.23.012.941 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.23.012.943 I srv  get_availabl: updating prompt cache
+0.23.012.949 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.23.012.953 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.23.012.954 I srv  get_availabl: prompt cache update took 0.01 ms
+0.23.013.012 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.26.234.628 I slot print_timing: id  0 | task 0 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.22 s / 1271.42 tokens per second
+0.27.896.790 I slot print_timing: id  0 | task 0 | prompt processing, n_tokens =   6011, progress = 0.85, t =   4.88 s / 1230.81 tokens per second
+0.27.944.722 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+0.28.834.529 I slot print_timing: id  0 | task 0 | prompt processing, n_tokens =   7035, progress = 1.00, t =   5.82 s / 1208.45 tokens per second
+0.28.892.962 I slot create_check: id  0 | task 0 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.29.188.213 I slot print_timing: id  0 | task 0 | 
+prompt eval time =    5911.23 ms /  7039 tokens (    0.84 ms per token,  1190.78 tokens per second)
+       eval time =     263.95 ms /    16 tokens (   16.50 ms per token,    60.62 tokens per second)
+      total time =    6175.18 ms /  7055 tokens
+0.29.188.522 I slot      release: id  0 | task 0 | stop processing: n_tokens = 7054, truncated = 0
+0.29.188.528 I srv  update_slots: all slots are idle
+0.29.206.128 I srv  params_from_: Chat format: peg-native
+0.29.206.239 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.994
+0.29.206.302 I slot launch_slot_: id  0 | task 21 | processing task, is_child = 0
+0.29.206.311 W slot update_slots: id  0 | task 21 | n_past = 7014, slot.prompt.tokens.size() = 7054, seq_id = 0, pos_min = 7053, n_swa = 0
+0.29.206.312 I slot update_slots: id  0 | task 21 | Checking checkpoint with [7034, 7034] against 7014...
+0.29.206.312 I slot update_slots: id  0 | task 21 | Checking checkpoint with [6010, 6010] against 7014...
+0.29.210.086 W slot update_slots: id  0 | task 21 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+0.29.210.090 W slot update_slots: id  0 | task 21 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+0.30.175.583 I slot create_check: id  0 | task 21 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.31.806.656 I slot print_timing: id  0 | task 21 | n_decoded =    100, tg =  62.48 t/s
+0.33.255.966 I slot print_timing: id  0 | task 21 | 
+prompt eval time =     999.89 ms /  1028 tokens (    0.97 ms per token,  1028.11 tokens per second)
+       eval time =    3049.75 ms /   192 tokens (   15.88 ms per token,    62.96 tokens per second)
+      total time =    4049.65 ms /  1220 tokens
+0.33.256.219 I slot      release: id  0 | task 21 | stop processing: n_tokens = 7230, truncated = 0
+0.33.256.240 I srv  update_slots: all slots are idle
+0.33.273.500 I srv  params_from_: Chat format: peg-native
+0.33.273.615 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.974
+0.33.273.668 I slot launch_slot_: id  0 | task 215 | processing task, is_child = 0
+0.33.273.676 W slot update_slots: id  0 | task 215 | erased invalidated context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.33.275.258 W slot update_slots: id  0 | task 215 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.625.268 I slot print_timing: id  0 | task 215 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.35 s / 1222.11 tokens per second
+0.38.375.032 I slot print_timing: id  0 | task 215 | prompt processing, n_tokens =   6011, progress = 0.85, t =   5.10 s / 1178.32 tokens per second
+0.38.426.513 I slot create_check: id  0 | task 215 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+0.39.343.585 I slot print_timing: id  0 | task 215 | prompt processing, n_tokens =   7035, progress = 1.00, t =   6.07 s / 1159.00 tokens per second
+0.39.401.811 I slot create_check: id  0 | task 215 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.41.031.975 I slot print_timing: id  0 | task 215 | n_decoded =    100, tg =  62.51 t/s
+0.42.481.538 I slot print_timing: id  0 | task 215 | 
+prompt eval time =    6158.66 ms /  7039 tokens (    0.87 ms per token,  1142.94 tokens per second)
+       eval time =    3049.19 ms /   192 tokens (   15.88 ms per token,    62.97 tokens per second)
+      total time =    9207.85 ms /  7231 tokens
+0.42.481.799 I slot      release: id  0 | task 215 | stop processing: n_tokens = 7230, truncated = 0
+0.42.481.812 I srv  update_slots: all slots are idle
+0.42.499.185 I srv  params_from_: Chat format: peg-native
+0.42.499.292 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.970
+0.42.499.320 I slot launch_slot_: id  0 | task 412 | processing task, is_child = 0
+0.42.499.327 W slot update_slots: id  0 | task 412 | n_past = 7014, slot.prompt.tokens.size() = 7230, seq_id = 0, pos_min = 7229, n_swa = 0
+0.42.499.328 I slot update_slots: id  0 | task 412 | Checking checkpoint with [7034, 7034] against 7014...
+0.42.499.328 I slot update_slots: id  0 | task 412 | Checking checkpoint with [6010, 6010] against 7014...
+0.42.502.995 W slot update_slots: id  0 | task 412 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+0.42.502.999 W slot update_slots: id  0 | task 412 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+0.43.471.709 I slot create_check: id  0 | task 412 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.43.764.311 I slot print_timing: id  0 | task 412 | 
+prompt eval time =    1002.90 ms /  1028 tokens (    0.98 ms per token,  1025.03 tokens per second)
+       eval time =     262.06 ms /    16 tokens (   16.38 ms per token,    61.05 tokens per second)
+      total time =    1264.96 ms /  1044 tokens
+0.43.764.770 I slot      release: id  0 | task 412 | stop processing: n_tokens = 7054, truncated = 0
+0.43.764.786 I srv  update_slots: all slots are idle
+0.43.792.285 I srv  params_from_: Chat format: peg-native
+0.43.792.395 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.994
+0.43.792.452 I slot launch_slot_: id  0 | task 430 | processing task, is_child = 0
+0.43.792.461 W slot update_slots: id  0 | task 430 | n_past = 7014, slot.prompt.tokens.size() = 7054, seq_id = 0, pos_min = 7053, n_swa = 0
+0.43.792.462 I slot update_slots: id  0 | task 430 | Checking checkpoint with [7034, 7034] against 7014...
+0.43.792.462 I slot update_slots: id  0 | task 430 | Checking checkpoint with [6010, 6010] against 7014...
+0.43.796.220 W slot update_slots: id  0 | task 430 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+0.43.796.224 W slot update_slots: id  0 | task 430 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+0.44.764.733 I slot create_check: id  0 | task 430 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.46.401.785 I slot print_timing: id  0 | task 430 | n_decoded =    100, tg =  62.25 t/s
+0.47.852.134 I slot print_timing: id  0 | task 430 | 
+prompt eval time =    1002.97 ms /  1028 tokens (    0.98 ms per token,  1024.96 tokens per second)
+       eval time =    3056.70 ms /   192 tokens (   15.92 ms per token,    62.81 tokens per second)
+      total time =    4059.66 ms /  1220 tokens
+0.47.852.392 I slot      release: id  0 | task 430 | stop processing: n_tokens = 7230, truncated = 0
+0.47.852.409 I srv  update_slots: all slots are idle
+0.47.869.577 I srv  params_from_: Chat format: peg-native
+0.47.869.673 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.974
+0.47.869.703 I slot launch_slot_: id  0 | task 624 | processing task, is_child = 0
+0.47.869.708 W slot update_slots: id  0 | task 624 | erased invalidated context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.47.871.268 W slot update_slots: id  0 | task 624 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.51.222.915 I slot print_timing: id  0 | task 624 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.35 s / 1221.52 tokens per second
+0.52.975.285 I slot print_timing: id  0 | task 624 | prompt processing, n_tokens =   6011, progress = 0.85, t =   5.11 s / 1177.34 tokens per second
+0.53.025.961 I slot create_check: id  0 | task 624 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+0.53.946.619 I slot print_timing: id  0 | task 624 | prompt processing, n_tokens =   7035, progress = 1.00, t =   6.08 s / 1157.66 tokens per second
+0.54.004.891 I slot create_check: id  0 | task 624 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.55.641.849 I slot print_timing: id  0 | task 624 | n_decoded =    100, tg =  62.25 t/s
+0.57.095.298 I slot print_timing: id  0 | task 624 | 
+prompt eval time =    6165.72 ms /  7039 tokens (    0.88 ms per token,  1141.64 tokens per second)
+       eval time =    3059.86 ms /   192 tokens (   15.94 ms per token,    62.75 tokens per second)
+      total time =    9225.58 ms /  7231 tokens
+0.57.095.560 I slot      release: id  0 | task 624 | stop processing: n_tokens = 7230, truncated = 0
+0.57.095.573 I srv  update_slots: all slots are idle
+0.57.112.963 I srv  params_from_: Chat format: peg-native
+0.57.113.061 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.970
+0.57.113.099 I slot launch_slot_: id  0 | task 821 | processing task, is_child = 0
+0.57.113.105 W slot update_slots: id  0 | task 821 | n_past = 7014, slot.prompt.tokens.size() = 7230, seq_id = 0, pos_min = 7229, n_swa = 0
+0.57.113.106 I slot update_slots: id  0 | task 821 | Checking checkpoint with [7034, 7034] against 7014...
+0.57.113.106 I slot update_slots: id  0 | task 821 | Checking checkpoint with [6010, 6010] against 7014...
+0.57.116.771 W slot update_slots: id  0 | task 821 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+0.57.116.773 W slot update_slots: id  0 | task 821 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+0.58.086.071 I slot create_check: id  0 | task 821 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+0.58.378.533 I slot print_timing: id  0 | task 821 | 
+prompt eval time =    1003.26 ms /  1028 tokens (    0.98 ms per token,  1024.66 tokens per second)
+       eval time =     262.14 ms /    16 tokens (   16.38 ms per token,    61.04 tokens per second)
+      total time =    1265.40 ms /  1044 tokens
+0.58.378.987 I slot      release: id  0 | task 821 | stop processing: n_tokens = 7054, truncated = 0
+0.58.379.007 I srv  update_slots: all slots are idle
+0.58.404.936 I srv  params_from_: Chat format: peg-native
+0.58.405.048 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.994
+0.58.405.100 I slot launch_slot_: id  0 | task 839 | processing task, is_child = 0
+0.58.405.109 W slot update_slots: id  0 | task 839 | n_past = 7014, slot.prompt.tokens.size() = 7054, seq_id = 0, pos_min = 7053, n_swa = 0
+0.58.405.109 I slot update_slots: id  0 | task 839 | Checking checkpoint with [7034, 7034] against 7014...
+0.58.405.110 I slot update_slots: id  0 | task 839 | Checking checkpoint with [6010, 6010] against 7014...
+0.58.408.850 W slot update_slots: id  0 | task 839 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+0.58.408.855 W slot update_slots: id  0 | task 839 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+0.59.378.102 I slot create_check: id  0 | task 839 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.01.014.891 I slot print_timing: id  0 | task 839 | n_decoded =    100, tg =  62.26 t/s
+1.02.468.788 I slot print_timing: id  0 | task 839 | 
+prompt eval time =    1003.48 ms /  1028 tokens (    0.98 ms per token,  1024.43 tokens per second)
+       eval time =    3060.19 ms /   192 tokens (   15.94 ms per token,    62.74 tokens per second)
+      total time =    4063.67 ms /  1220 tokens
+1.02.469.046 I slot      release: id  0 | task 839 | stop processing: n_tokens = 7230, truncated = 0
+1.02.469.057 I srv  update_slots: all slots are idle
+1.02.486.085 I srv  params_from_: Chat format: peg-native
+1.02.486.197 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.974
+1.02.486.250 I slot launch_slot_: id  0 | task 1033 | processing task, is_child = 0
+1.02.486.266 W slot update_slots: id  0 | task 1033 | erased invalidated context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.02.488.866 W slot update_slots: id  0 | task 1033 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.05.842.274 I slot print_timing: id  0 | task 1033 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.36 s / 1220.51 tokens per second
+1.07.594.394 I slot print_timing: id  0 | task 1033 | prompt processing, n_tokens =   6011, progress = 0.85, t =   5.11 s / 1176.75 tokens per second
+1.07.645.015 I slot create_check: id  0 | task 1033 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+1.08.566.063 I slot print_timing: id  0 | task 1033 | prompt processing, n_tokens =   7035, progress = 1.00, t =   6.08 s / 1157.11 tokens per second
+1.08.624.878 I slot create_check: id  0 | task 1033 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.10.262.102 I slot print_timing: id  0 | task 1033 | n_decoded =    100, tg =  62.24 t/s
+1.11.717.464 I slot print_timing: id  0 | task 1033 | 
+prompt eval time =    6169.04 ms /  7039 tokens (    0.88 ms per token,  1141.02 tokens per second)
+       eval time =    3062.15 ms /   192 tokens (   15.95 ms per token,    62.70 tokens per second)
+      total time =    9231.19 ms /  7231 tokens
+1.11.717.716 I slot      release: id  0 | task 1033 | stop processing: n_tokens = 7230, truncated = 0
+1.11.717.731 I srv  update_slots: all slots are idle
+1.11.734.739 I srv  params_from_: Chat format: peg-native
+1.11.734.856 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.970
+1.11.734.912 I slot launch_slot_: id  0 | task 1230 | processing task, is_child = 0
+1.11.734.923 W slot update_slots: id  0 | task 1230 | n_past = 7014, slot.prompt.tokens.size() = 7230, seq_id = 0, pos_min = 7229, n_swa = 0
+1.11.734.923 I slot update_slots: id  0 | task 1230 | Checking checkpoint with [7034, 7034] against 7014...
+1.11.734.924 I slot update_slots: id  0 | task 1230 | Checking checkpoint with [6010, 6010] against 7014...
+1.11.738.681 W slot update_slots: id  0 | task 1230 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+1.11.738.683 W slot update_slots: id  0 | task 1230 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+1.12.708.145 I slot create_check: id  0 | task 1230 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.13.000.920 I slot print_timing: id  0 | task 1230 | 
+prompt eval time =    1003.98 ms /  1028 tokens (    0.98 ms per token,  1023.92 tokens per second)
+       eval time =     262.00 ms /    16 tokens (   16.37 ms per token,    61.07 tokens per second)
+      total time =    1265.98 ms /  1044 tokens
+1.13.001.384 I slot      release: id  0 | task 1230 | stop processing: n_tokens = 7054, truncated = 0
+1.13.001.412 I srv  update_slots: all slots are idle
+1.13.029.139 I srv  params_from_: Chat format: peg-native
+1.13.029.237 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.994
+1.13.029.274 I slot launch_slot_: id  0 | task 1248 | processing task, is_child = 0
+1.13.029.280 W slot update_slots: id  0 | task 1248 | n_past = 7014, slot.prompt.tokens.size() = 7054, seq_id = 0, pos_min = 7053, n_swa = 0
+1.13.029.280 I slot update_slots: id  0 | task 1248 | Checking checkpoint with [7034, 7034] against 7014...
+1.13.029.281 I slot update_slots: id  0 | task 1248 | Checking checkpoint with [6010, 6010] against 7014...
+1.13.032.947 W slot update_slots: id  0 | task 1248 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+1.13.032.950 W slot update_slots: id  0 | task 1248 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+1.14.002.308 I slot create_check: id  0 | task 1248 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.15.640.869 I slot print_timing: id  0 | task 1248 | n_decoded =    100, tg =  62.18 t/s
+1.17.096.125 I slot print_timing: id  0 | task 1248 | 
+prompt eval time =    1003.44 ms /  1028 tokens (    0.98 ms per token,  1024.48 tokens per second)
+       eval time =    3063.39 ms /   192 tokens (   15.96 ms per token,    62.68 tokens per second)
+      total time =    4066.83 ms /  1220 tokens
+1.17.096.385 I slot      release: id  0 | task 1248 | stop processing: n_tokens = 7230, truncated = 0
+1.17.096.398 I srv  update_slots: all slots are idle
+1.17.113.652 I srv  params_from_: Chat format: peg-native
+1.17.113.773 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.974
+1.17.113.827 I slot launch_slot_: id  0 | task 1442 | processing task, is_child = 0
+1.17.113.835 W slot update_slots: id  0 | task 1442 | erased invalidated context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.17.115.551 W slot update_slots: id  0 | task 1442 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.20.471.772 I slot print_timing: id  0 | task 1442 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.36 s / 1219.80 tokens per second
+1.22.222.821 I slot print_timing: id  0 | task 1442 | prompt processing, n_tokens =   6011, progress = 0.85, t =   5.11 s / 1176.56 tokens per second
+1.22.273.999 I slot create_check: id  0 | task 1442 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+1.23.196.105 I slot print_timing: id  0 | task 1442 | prompt processing, n_tokens =   7035, progress = 1.00, t =   6.08 s / 1156.64 tokens per second
+1.23.254.243 I slot create_check: id  0 | task 1442 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.24.893.156 I slot print_timing: id  0 | task 1442 | n_decoded =    100, tg =  62.17 t/s
+1.26.350.431 I slot print_timing: id  0 | task 1442 | 
+prompt eval time =    6170.80 ms /  7039 tokens (    0.88 ms per token,  1140.70 tokens per second)
+       eval time =    3065.79 ms /   192 tokens (   15.97 ms per token,    62.63 tokens per second)
+      total time =    9236.59 ms /  7231 tokens
+1.26.350.694 I slot      release: id  0 | task 1442 | stop processing: n_tokens = 7230, truncated = 0
+1.26.350.710 I srv  update_slots: all slots are idle
+1.26.367.931 I srv  params_from_: Chat format: peg-native
+1.26.368.047 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.970
+1.26.368.104 I slot launch_slot_: id  0 | task 1639 | processing task, is_child = 0
+1.26.368.113 W slot update_slots: id  0 | task 1639 | n_past = 7014, slot.prompt.tokens.size() = 7230, seq_id = 0, pos_min = 7229, n_swa = 0
+1.26.368.113 I slot update_slots: id  0 | task 1639 | Checking checkpoint with [7034, 7034] against 7014...
+1.26.368.114 I slot update_slots: id  0 | task 1639 | Checking checkpoint with [6010, 6010] against 7014...
+1.26.371.873 W slot update_slots: id  0 | task 1639 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+1.26.371.876 W slot update_slots: id  0 | task 1639 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+1.27.342.456 I slot create_check: id  0 | task 1639 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.27.635.050 I slot print_timing: id  0 | task 1639 | 
+prompt eval time =    1004.71 ms /  1028 tokens (    0.98 ms per token,  1023.18 tokens per second)
+       eval time =     262.20 ms /    16 tokens (   16.39 ms per token,    61.02 tokens per second)
+      total time =    1266.92 ms /  1044 tokens
+1.27.635.512 I slot      release: id  0 | task 1639 | stop processing: n_tokens = 7054, truncated = 0
+1.27.635.532 I srv  update_slots: all slots are idle
+1.27.666.084 I srv  params_from_: Chat format: peg-native
+1.27.666.197 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.996 (> 0.100 thold), f_keep = 0.994
+1.27.666.254 I slot launch_slot_: id  0 | task 1657 | processing task, is_child = 0
+1.27.666.272 W slot update_slots: id  0 | task 1657 | n_past = 7014, slot.prompt.tokens.size() = 7054, seq_id = 0, pos_min = 7053, n_swa = 0
+1.27.666.273 I slot update_slots: id  0 | task 1657 | Checking checkpoint with [7034, 7034] against 7014...
+1.27.666.273 I slot update_slots: id  0 | task 1657 | Checking checkpoint with [6010, 6010] against 7014...
+1.27.670.025 W slot update_slots: id  0 | task 1657 | restored context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_past = 6011, size = 62.813 MiB)
+1.27.670.030 W slot update_slots: id  0 | task 1657 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 6011, size = 62.813 MiB)
+1.28.641.307 I slot create_check: id  0 | task 1657 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.30.279.245 I slot print_timing: id  0 | task 1657 | n_decoded =    100, tg =  62.21 t/s
+1.31.732.686 I slot print_timing: id  0 | task 1657 | 
+prompt eval time =    1005.61 ms /  1028 tokens (    0.98 ms per token,  1022.27 tokens per second)
+       eval time =    3060.81 ms /   192 tokens (   15.94 ms per token,    62.73 tokens per second)
+      total time =    4066.42 ms /  1220 tokens
+1.31.732.942 I slot      release: id  0 | task 1657 | stop processing: n_tokens = 7230, truncated = 0
+1.31.732.953 I srv  update_slots: all slots are idle
+1.31.750.071 I srv  params_from_: Chat format: peg-native
+1.31.750.187 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 1.000 (> 0.100 thold), f_keep = 0.974
+1.31.750.241 I slot launch_slot_: id  0 | task 1851 | processing task, is_child = 0
+1.31.750.248 W slot update_slots: id  0 | task 1851 | erased invalidated context checkpoint (pos_min = 6010, pos_max = 6010, n_tokens = 6011, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.31.751.779 W slot update_slots: id  0 | task 1851 | erased invalidated context checkpoint (pos_min = 7034, pos_max = 7034, n_tokens = 7035, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+1.35.110.825 I slot print_timing: id  0 | task 1851 | prompt processing, n_tokens =   4096, progress = 0.58, t =   3.36 s / 1218.84 tokens per second
+1.36.865.814 I slot print_timing: id  0 | task 1851 | prompt processing, n_tokens =   6011, progress = 0.85, t =   5.12 s / 1175.04 tokens per second
+1.36.917.039 I slot create_check: id  0 | task 1851 | created context checkpoint 1 of 32 (pos_min = 6010, pos_max = 6010, n_tokens = 6011, size = 62.813 MiB)
+1.37.839.294 I slot print_timing: id  0 | task 1851 | prompt processing, n_tokens =   7035, progress = 1.00, t =   6.09 s / 1155.36 tokens per second
+1.37.897.427 I slot create_check: id  0 | task 1851 | created context checkpoint 2 of 32 (pos_min = 7034, pos_max = 7034, n_tokens = 7035, size = 62.813 MiB)
+1.39.536.081 I slot print_timing: id  0 | task 1851 | n_decoded =    100, tg =  62.18 t/s
+1.40.995.612 I slot print_timing: id  0 | task 1851 | 
+prompt eval time =    6177.63 ms /  7039 tokens (    0.88 ms per token,  1139.43 tokens per second)
+       eval time =    3067.72 ms /   192 tokens (   15.98 ms per token,    62.59 tokens per second)
+      total time =    9245.35 ms /  7231 tokens
+1.40.995.868 I slot      release: id  0 | task 1851 | stop processing: n_tokens = 7230, truncated = 0
+1.40.995.883 I srv  update_slots: all slots are idle
+1.40.996.543 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-c1-probe.log

@@ -0,0 +1,286 @@
+0.00.113.666 W Setting 'enable_thinking' via --chat-template-kwargs is deprecated. Use --reasoning on / --reasoning off instead.
+0.00.124.179 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.124.183 I device_info:
+0.00.124.292 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 123913 MiB free)
+0.00.124.424 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.124.432 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.124.514 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.124.572 I srv          init: running without SSL
+0.00.124.596 I srv          init: using 31 threads for HTTP server
+0.00.124.597 I srv          init: the WebUI is disabled
+0.00.124.672 I srv         start: binding port with default address family
+0.00.125.865 I srv          main: loading model
+0.00.125.873 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.172.940 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.22.203.780 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.22.400.933 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.718.577 I srv    load_model: initializing slots, n_slots = 1
+0.22.944.184 W srv    load_model: speculative decoding will use checkpoints
+0.22.944.193 W common_speculative_init: no implementations specified for speculative decoding
+0.22.944.194 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.22.944.259 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.22.944.261 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.22.944.283 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.22.956.362 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+'
+0.22.964.154 I srv          init: init: chat template, thinking = 1
+0.22.964.188 I srv          main: model loaded
+0.22.964.191 I srv          main: server is listening on http://127.0.0.1:18652
+0.22.964.207 I srv  update_slots: all slots are idle
+0.24.203.095 I srv  params_from_: Chat format: peg-native
+0.24.203.661 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.203.663 I srv  get_availabl: updating prompt cache
+0.24.203.670 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.203.676 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.203.678 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.203.955 I reasoning-budget: activated, budget=2147483647 tokens
+0.24.203.957 I reasoning-budget: deactivated (natural end)
+0.24.203.972 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.920.604 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.25.129.936 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     758.65 ms /   427 tokens (    1.78 ms per token,   562.84 tokens per second)
+       eval time =     167.29 ms /     4 tokens (   41.82 ms per token,    23.91 tokens per second)
+      total time =     925.94 ms /   431 tokens
+0.25.130.011 I slot      release: id  0 | task 0 | stop processing: n_tokens = 430, truncated = 0
+0.25.130.019 I srv  update_slots: all slots are idle
+0.25.185.780 I srv  params_from_: Chat format: peg-native
+0.25.186.376 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.942
+0.25.186.613 I reasoning-budget: activated, budget=2147483647 tokens
+0.25.186.618 I reasoning-budget: deactivated (natural end)
+0.25.186.779 I slot launch_slot_: id  0 | task 6 | processing task, is_child = 0
+0.25.186.816 W slot update_slots: id  0 | task 6 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.25.186.819 I slot update_slots: id  0 | task 6 | Checking checkpoint with [422, 422] against 405...
+0.25.186.822 W slot update_slots: id  0 | task 6 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.25.186.829 W slot update_slots: id  0 | task 6 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.25.897.246 I slot create_check: id  0 | task 6 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.25.983.622 I slot print_timing: id  0 | task 6 | 
+prompt eval time =     763.27 ms /   421 tokens (    1.81 ms per token,   551.57 tokens per second)
+       eval time =      33.53 ms /     2 tokens (   16.77 ms per token,    59.64 tokens per second)
+      total time =     796.80 ms /   423 tokens
+0.25.983.694 I slot      release: id  0 | task 6 | stop processing: n_tokens = 422, truncated = 0
+0.25.983.724 I srv  update_slots: all slots are idle
+0.26.010.893 I srv  params_from_: Chat format: peg-native
+0.26.011.386 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.960
+0.26.011.582 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.011.584 I reasoning-budget: deactivated (natural end)
+0.26.011.615 I slot launch_slot_: id  0 | task 10 | processing task, is_child = 0
+0.26.011.627 W slot update_slots: id  0 | task 10 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.26.011.628 I slot update_slots: id  0 | task 10 | Checking checkpoint with [416, 416] against 405...
+0.26.011.629 W slot update_slots: id  0 | task 10 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.26.011.634 W slot update_slots: id  0 | task 10 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.26.506.256 I slot create_check: id  0 | task 10 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.27.421.771 I slot print_timing: id  0 | task 10 | 
+prompt eval time =     545.66 ms /   424 tokens (    1.29 ms per token,   777.04 tokens per second)
+       eval time =     864.45 ms /    39 tokens (   22.17 ms per token,    45.12 tokens per second)
+      total time =    1410.11 ms /   463 tokens
+0.27.421.958 I slot      release: id  0 | task 10 | stop processing: n_tokens = 462, truncated = 0
+0.27.422.017 I srv  update_slots: all slots are idle
+0.27.483.797 I srv  params_from_: Chat format: peg-native
+0.27.486.309 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.951 (> 0.100 thold), f_keep = 0.879
+0.27.486.919 I reasoning-budget: activated, budget=2147483647 tokens
+0.27.486.923 I reasoning-budget: deactivated (natural end)
+0.27.487.006 I slot launch_slot_: id  0 | task 51 | processing task, is_child = 0
+0.27.487.029 W slot update_slots: id  0 | task 51 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.27.487.032 I slot update_slots: id  0 | task 51 | Checking checkpoint with [419, 419] against 406...
+0.27.487.034 W slot update_slots: id  0 | task 51 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.27.487.041 W slot update_slots: id  0 | task 51 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.076.337 I slot create_check: id  0 | task 51 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.28.262.498 I slot print_timing: id  0 | task 51 | 
+prompt eval time =     638.39 ms /   427 tokens (    1.50 ms per token,   668.86 tokens per second)
+       eval time =     137.05 ms /     4 tokens (   34.26 ms per token,    29.19 tokens per second)
+      total time =     775.45 ms /   431 tokens
+0.28.262.688 I slot      release: id  0 | task 51 | stop processing: n_tokens = 430, truncated = 0
+0.28.262.742 I srv  update_slots: all slots are idle
+0.28.315.880 I srv  params_from_: Chat format: peg-native
+0.28.318.140 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.942
+0.28.318.686 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.318.690 I reasoning-budget: deactivated (natural end)
+0.28.318.777 I slot launch_slot_: id  0 | task 57 | processing task, is_child = 0
+0.28.318.800 W slot update_slots: id  0 | task 57 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.28.318.803 I slot update_slots: id  0 | task 57 | Checking checkpoint with [422, 422] against 405...
+0.28.318.805 W slot update_slots: id  0 | task 57 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.318.810 W slot update_slots: id  0 | task 57 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.907.669 I slot create_check: id  0 | task 57 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.28.982.494 I slot print_timing: id  0 | task 57 | 
+prompt eval time =     636.88 ms /   421 tokens (    1.51 ms per token,   661.04 tokens per second)
+       eval time =      26.81 ms /     2 tokens (   13.41 ms per token,    74.60 tokens per second)
+      total time =     663.69 ms /   423 tokens
+0.28.982.572 I slot      release: id  0 | task 57 | stop processing: n_tokens = 422, truncated = 0
+0.28.982.598 I srv  update_slots: all slots are idle
+0.29.009.222 I srv  params_from_: Chat format: peg-native
+0.29.011.294 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.960
+0.29.011.848 I reasoning-budget: activated, budget=2147483647 tokens
+0.29.011.851 I reasoning-budget: deactivated (natural end)
+0.29.011.922 I slot launch_slot_: id  0 | task 61 | processing task, is_child = 0
+0.29.011.941 W slot update_slots: id  0 | task 61 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.29.011.943 I slot update_slots: id  0 | task 61 | Checking checkpoint with [416, 416] against 405...
+0.29.011.945 W slot update_slots: id  0 | task 61 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.29.011.949 W slot update_slots: id  0 | task 61 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.29.554.822 I slot create_check: id  0 | task 61 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.30.389.784 I slot print_timing: id  0 | task 61 | 
+prompt eval time =     589.42 ms /   424 tokens (    1.39 ms per token,   719.35 tokens per second)
+       eval time =     788.41 ms /    39 tokens (   20.22 ms per token,    49.47 tokens per second)
+      total time =    1377.83 ms /   463 tokens
+0.30.389.858 I slot      release: id  0 | task 61 | stop processing: n_tokens = 462, truncated = 0
+0.30.389.883 I srv  update_slots: all slots are idle
+0.30.401.499 I srv  params_from_: Chat format: peg-native
+0.30.401.853 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.879
+0.30.402.127 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.402.129 I reasoning-budget: deactivated (natural end)
+0.30.402.192 I slot launch_slot_: id  0 | task 102 | processing task, is_child = 0
+0.30.402.206 W slot update_slots: id  0 | task 102 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.30.402.208 I slot update_slots: id  0 | task 102 | Checking checkpoint with [419, 419] against 406...
+0.30.402.209 W slot update_slots: id  0 | task 102 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.402.212 W slot update_slots: id  0 | task 102 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.923.747 I slot create_check: id  0 | task 102 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.31.379.211 I slot print_timing: id  0 | task 102 | 
+prompt eval time =     572.31 ms /   425 tokens (    1.35 ms per token,   742.61 tokens per second)
+       eval time =     404.68 ms /    17 tokens (   23.80 ms per token,    42.01 tokens per second)
+      total time =     976.98 ms /   442 tokens
+0.31.379.332 I slot      release: id  0 | task 102 | stop processing: n_tokens = 441, truncated = 0
+0.31.379.373 I srv  update_slots: all slots are idle
+0.31.405.914 I srv  params_from_: Chat format: peg-native
+0.31.406.236 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.918
+0.31.406.397 I reasoning-budget: activated, budget=2147483647 tokens
+0.31.406.398 I reasoning-budget: deactivated (natural end)
+0.31.406.427 I slot launch_slot_: id  0 | task 121 | processing task, is_child = 0
+0.31.406.435 W slot update_slots: id  0 | task 121 | n_past = 405, slot.prompt.tokens.size() = 441, seq_id = 0, pos_min = 440, n_swa = 0
+0.31.406.436 I slot update_slots: id  0 | task 121 | Checking checkpoint with [420, 420] against 405...
+0.31.406.437 W slot update_slots: id  0 | task 121 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.31.406.439 W slot update_slots: id  0 | task 121 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.31.861.565 I slot create_check: id  0 | task 121 | created context checkpoint 1 of 32 (pos_min = 414, pos_max = 414, n_tokens = 415, size = 62.813 MiB)
+0.32.160.128 I slot print_timing: id  0 | task 121 | 
+prompt eval time =     491.31 ms /   419 tokens (    1.17 ms per token,   852.82 tokens per second)
+       eval time =     262.37 ms /    12 tokens (   21.86 ms per token,    45.74 tokens per second)
+      total time =     753.68 ms /   431 tokens
+0.32.160.211 I slot      release: id  0 | task 121 | stop processing: n_tokens = 430, truncated = 0
+0.32.160.241 I srv  update_slots: all slots are idle
+0.32.209.194 I srv  params_from_: Chat format: peg-native
+0.32.209.712 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.960 (> 0.100 thold), f_keep = 0.942
+0.32.209.953 I reasoning-budget: activated, budget=2147483647 tokens
+0.32.209.954 I reasoning-budget: deactivated (natural end)
+0.32.209.994 I slot launch_slot_: id  0 | task 135 | processing task, is_child = 0
+0.32.210.006 W slot update_slots: id  0 | task 135 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.32.210.007 I slot update_slots: id  0 | task 135 | Checking checkpoint with [414, 414] against 405...
+0.32.210.008 W slot update_slots: id  0 | task 135 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.32.210.011 W slot update_slots: id  0 | task 135 | erased invalidated context checkpoint (pos_min = 414, pos_max = 414, n_tokens = 415, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.32.786.978 I slot create_check: id  0 | task 135 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.33.839.717 I slot print_timing: id  0 | task 135 | 
+prompt eval time =     614.39 ms /   422 tokens (    1.46 ms per token,   686.86 tokens per second)
+       eval time =    1015.30 ms /    53 tokens (   19.16 ms per token,    52.20 tokens per second)
+      total time =    1629.69 ms /   475 tokens
+0.33.839.794 I slot      release: id  0 | task 135 | stop processing: n_tokens = 474, truncated = 0
+0.33.839.825 I srv  update_slots: all slots are idle
+0.33.879.962 I srv  params_from_: Chat format: peg-native
+0.33.880.516 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.857
+0.33.880.743 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.880.744 I reasoning-budget: deactivated (natural end)
+0.33.880.792 I slot launch_slot_: id  0 | task 190 | processing task, is_child = 0
+0.33.880.805 W slot update_slots: id  0 | task 190 | n_past = 406, slot.prompt.tokens.size() = 474, seq_id = 0, pos_min = 473, n_swa = 0
+0.33.880.806 I slot update_slots: id  0 | task 190 | Checking checkpoint with [417, 417] against 406...
+0.33.880.807 W slot update_slots: id  0 | task 190 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.33.880.810 W slot update_slots: id  0 | task 190 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.34.423.705 I slot create_check: id  0 | task 190 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.34.628.920 I slot print_timing: id  0 | task 190 | 
+prompt eval time =     583.31 ms /   424 tokens (    1.38 ms per token,   726.89 tokens per second)
+       eval time =     164.80 ms /     7 tokens (   23.54 ms per token,    42.48 tokens per second)
+      total time =     748.11 ms /   431 tokens
+0.34.628.991 I slot      release: id  0 | task 190 | stop processing: n_tokens = 430, truncated = 0
+0.34.629.018 I srv  update_slots: all slots are idle
+0.34.642.421 I srv  params_from_: Chat format: peg-native
+0.34.642.945 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.969 (> 0.100 thold), f_keep = 0.942
+0.34.643.148 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.643.150 I reasoning-budget: deactivated (natural end)
+0.34.643.185 I slot launch_slot_: id  0 | task 199 | processing task, is_child = 0
+0.34.643.195 W slot update_slots: id  0 | task 199 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.34.643.196 I slot update_slots: id  0 | task 199 | Checking checkpoint with [419, 419] against 405...
+0.34.643.197 W slot update_slots: id  0 | task 199 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.34.643.199 W slot update_slots: id  0 | task 199 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.154.839 I slot create_check: id  0 | task 199 | created context checkpoint 1 of 32 (pos_min = 413, pos_max = 413, n_tokens = 414, size = 62.813 MiB)
+0.35.329.011 I slot print_timing: id  0 | task 199 | 
+prompt eval time =     563.78 ms /   418 tokens (    1.35 ms per token,   741.42 tokens per second)
+       eval time =     122.02 ms /     5 tokens (   24.40 ms per token,    40.98 tokens per second)
+      total time =     685.80 ms /   423 tokens
+0.35.329.095 I slot      release: id  0 | task 199 | stop processing: n_tokens = 422, truncated = 0
+0.35.329.136 I srv  update_slots: all slots are idle
+0.35.355.115 I srv  params_from_: Chat format: peg-native
+0.35.355.649 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.960
+0.35.356.326 I reasoning-budget: activated, budget=2147483647 tokens
+0.35.356.340 I reasoning-budget: deactivated (natural end)
+0.35.356.552 I slot launch_slot_: id  0 | task 206 | processing task, is_child = 0
+0.35.356.603 W slot update_slots: id  0 | task 206 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.35.356.611 I slot update_slots: id  0 | task 206 | Checking checkpoint with [413, 413] against 405...
+0.35.356.616 W slot update_slots: id  0 | task 206 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.35.356.634 W slot update_slots: id  0 | task 206 | erased invalidated context checkpoint (pos_min = 413, pos_max = 413, n_tokens = 414, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.935.879 I slot create_check: id  0 | task 206 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.36.896.774 I slot print_timing: id  0 | task 206 | 
+prompt eval time =     625.99 ms /   421 tokens (    1.49 ms per token,   672.53 tokens per second)
+       eval time =     914.17 ms /    42 tokens (   21.77 ms per token,    45.94 tokens per second)
+      total time =    1540.17 ms /   463 tokens
+0.36.896.862 I slot      release: id  0 | task 206 | stop processing: n_tokens = 462, truncated = 0
+0.36.896.896 I srv  update_slots: all slots are idle
+0.36.936.572 I srv  params_from_: Chat format: peg-native
+0.36.937.088 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.951 (> 0.100 thold), f_keep = 0.879
+0.36.937.669 I reasoning-budget: activated, budget=2147483647 tokens
+0.36.937.673 I reasoning-budget: deactivated (natural end)
+0.36.937.771 I slot launch_slot_: id  0 | task 250 | processing task, is_child = 0
+0.36.937.796 W slot update_slots: id  0 | task 250 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.36.937.799 I slot update_slots: id  0 | task 250 | Checking checkpoint with [416, 416] against 406...
+0.36.937.801 W slot update_slots: id  0 | task 250 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.36.937.807 W slot update_slots: id  0 | task 250 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.37.457.001 I slot create_check: id  0 | task 250 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.37.596.606 I slot print_timing: id  0 | task 250 | 
+prompt eval time =     561.66 ms /   427 tokens (    1.32 ms per token,   760.24 tokens per second)
+       eval time =      97.14 ms /     4 tokens (   24.28 ms per token,    41.18 tokens per second)
+      total time =     658.80 ms /   431 tokens
+0.37.596.716 I slot      release: id  0 | task 250 | stop processing: n_tokens = 430, truncated = 0
+0.37.596.750 I srv  update_slots: all slots are idle
+0.37.626.974 I srv  params_from_: Chat format: peg-native
+0.37.627.483 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.942
+0.37.627.753 I reasoning-budget: activated, budget=2147483647 tokens
+0.37.627.755 I reasoning-budget: deactivated (natural end)
+0.37.627.791 I slot launch_slot_: id  0 | task 256 | processing task, is_child = 0
+0.37.627.804 W slot update_slots: id  0 | task 256 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.37.627.806 I slot update_slots: id  0 | task 256 | Checking checkpoint with [422, 422] against 405...
+0.37.627.807 W slot update_slots: id  0 | task 256 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.37.627.811 W slot update_slots: id  0 | task 256 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.38.154.752 I slot create_check: id  0 | task 256 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.38.242.158 I slot print_timing: id  0 | task 256 | 
+prompt eval time =     577.92 ms /   421 tokens (    1.37 ms per token,   728.48 tokens per second)
+       eval time =      36.41 ms /     2 tokens (   18.21 ms per token,    54.92 tokens per second)
+      total time =     614.33 ms /   423 tokens
+0.38.242.323 I slot      release: id  0 | task 256 | stop processing: n_tokens = 422, truncated = 0
+0.38.242.358 I srv  update_slots: all slots are idle
+0.38.257.074 I srv  params_from_: Chat format: peg-native
+0.38.257.513 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.960
+0.38.257.760 I reasoning-budget: activated, budget=2147483647 tokens
+0.38.257.761 I reasoning-budget: deactivated (natural end)
+0.38.257.798 I slot launch_slot_: id  0 | task 260 | processing task, is_child = 0
+0.38.257.809 W slot update_slots: id  0 | task 260 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.38.257.809 I slot update_slots: id  0 | task 260 | Checking checkpoint with [416, 416] against 405...
+0.38.257.810 W slot update_slots: id  0 | task 260 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.38.257.813 W slot update_slots: id  0 | task 260 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.38.777.417 I slot create_check: id  0 | task 260 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.39.603.048 I slot print_timing: id  0 | task 260 | 
+prompt eval time =     562.75 ms /   424 tokens (    1.33 ms per token,   753.44 tokens per second)
+       eval time =     782.47 ms /    39 tokens (   20.06 ms per token,    49.84 tokens per second)
+      total time =    1345.22 ms /   463 tokens
+0.39.603.254 I slot      release: id  0 | task 260 | stop processing: n_tokens = 462, truncated = 0
+0.39.603.336 I srv  update_slots: all slots are idle
+0.39.604.550 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-c1.log

@@ -0,0 +1,302 @@
+0.00.120.044 W Setting 'enable_thinking' via --chat-template-kwargs is deprecated. Use --reasoning on / --reasoning off instead.
+0.00.163.630 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.163.643 I device_info:
+0.00.163.785 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 123858 MiB free)
+0.00.163.974 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.163.982 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.164.078 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.164.142 I srv          init: running without SSL
+0.00.164.166 I srv          init: using 31 threads for HTTP server
+0.00.164.167 I srv          init: the WebUI is disabled
+0.00.164.248 I srv         start: binding port with default address family
+0.00.165.500 I srv          main: loading model
+0.00.165.507 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.216.012 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.22.544.117 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.22.849.861 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.23.278.480 I srv    load_model: initializing slots, n_slots = 1
+0.23.567.593 W srv    load_model: speculative decoding will use checkpoints
+0.23.567.602 W common_speculative_init: no implementations specified for speculative decoding
+0.23.567.605 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.23.567.670 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.23.567.688 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.23.567.709 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.23.579.583 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+'
+0.23.587.209 I srv          init: init: chat template, thinking = 1
+0.23.587.253 I srv          main: model loaded
+0.23.587.256 I srv          main: server is listening on http://127.0.0.1:18600
+0.23.587.260 I srv  update_slots: all slots are idle
+0.24.570.949 I srv  params_from_: Chat format: peg-native
+0.24.571.406 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.571.411 I srv  get_availabl: updating prompt cache
+0.24.571.419 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.571.426 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.571.428 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.571.805 I reasoning-budget: activated, budget=2147483647 tokens
+0.24.571.827 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.25.235.195 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.25.806.187 I reasoning-budget: deactivated (natural end)
+0.26.650.056 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     715.70 ms /   422 tokens (    1.70 ms per token,   589.63 tokens per second)
+       eval time =    1362.48 ms /    60 tokens (   22.71 ms per token,    44.04 tokens per second)
+      total time =    2078.18 ms /   482 tokens
+0.26.650.247 I slot      release: id  0 | task 0 | stop processing: n_tokens = 481, truncated = 0
+0.26.650.264 I srv  update_slots: all slots are idle
+0.26.675.130 I srv  params_from_: Chat format: peg-native
+0.26.675.629 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.906 (> 0.100 thold), f_keep = 0.842
+0.26.676.106 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.676.196 I slot launch_slot_: id  0 | task 62 | processing task, is_child = 0
+0.26.676.216 W slot update_slots: id  0 | task 62 | n_past = 405, slot.prompt.tokens.size() = 481, seq_id = 0, pos_min = 480, n_swa = 0
+0.26.676.222 I slot update_slots: id  0 | task 62 | Checking checkpoint with [417, 417] against 405...
+0.26.676.224 W slot update_slots: id  0 | task 62 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.26.676.229 W slot update_slots: id  0 | task 62 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.27.231.790 I slot create_check: id  0 | task 62 | created context checkpoint 1 of 32 (pos_min = 442, pos_max = 442, n_tokens = 443, size = 62.813 MiB)
+0.27.733.232 I reasoning-budget: deactivated (natural end)
+0.28.618.139 I slot print_timing: id  0 | task 62 | 
+prompt eval time =     593.45 ms /   447 tokens (    1.33 ms per token,   753.22 tokens per second)
+       eval time =    1348.46 ms /    65 tokens (   20.75 ms per token,    48.20 tokens per second)
+      total time =    1941.91 ms /   512 tokens
+0.28.618.225 I slot      release: id  0 | task 62 | stop processing: n_tokens = 511, truncated = 0
+0.28.618.260 I srv  update_slots: all slots are idle
+0.28.632.505 I srv  params_from_: Chat format: peg-native
+0.28.632.871 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.793
+0.28.633.063 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.633.092 I slot launch_slot_: id  0 | task 129 | processing task, is_child = 0
+0.28.633.101 W slot update_slots: id  0 | task 129 | n_past = 405, slot.prompt.tokens.size() = 511, seq_id = 0, pos_min = 510, n_swa = 0
+0.28.633.102 I slot update_slots: id  0 | task 129 | Checking checkpoint with [442, 442] against 405...
+0.28.633.103 W slot update_slots: id  0 | task 129 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.633.104 W slot update_slots: id  0 | task 129 | erased invalidated context checkpoint (pos_min = 442, pos_max = 442, n_tokens = 443, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.29.161.759 I slot create_check: id  0 | task 129 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.29.478.654 I reasoning-budget: deactivated (natural end)
+0.30.331.330 I slot print_timing: id  0 | task 129 | 
+prompt eval time =     580.01 ms /   425 tokens (    1.36 ms per token,   732.75 tokens per second)
+       eval time =    1118.18 ms /    52 tokens (   21.50 ms per token,    46.50 tokens per second)
+      total time =    1698.19 ms /   477 tokens
+0.30.331.521 I slot      release: id  0 | task 129 | stop processing: n_tokens = 476, truncated = 0
+0.30.331.586 I srv  update_slots: all slots are idle
+0.30.386.522 I srv  params_from_: Chat format: peg-native
+0.30.388.735 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.851
+0.30.389.319 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.389.437 I slot launch_slot_: id  0 | task 183 | processing task, is_child = 0
+0.30.389.465 W slot update_slots: id  0 | task 183 | n_past = 405, slot.prompt.tokens.size() = 476, seq_id = 0, pos_min = 475, n_swa = 0
+0.30.389.469 I slot update_slots: id  0 | task 183 | Checking checkpoint with [420, 420] against 405...
+0.30.389.471 W slot update_slots: id  0 | task 183 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.389.478 W slot update_slots: id  0 | task 183 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.972.513 I slot create_check: id  0 | task 183 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.31.338.635 I reasoning-budget: deactivated (natural end)
+0.31.461.621 I slot print_timing: id  0 | task 183 | 
+prompt eval time =     630.63 ms /   425 tokens (    1.48 ms per token,   673.92 tokens per second)
+       eval time =     441.50 ms /    17 tokens (   25.97 ms per token,    38.51 tokens per second)
+      total time =    1072.13 ms /   442 tokens
+0.31.461.844 I slot      release: id  0 | task 183 | stop processing: n_tokens = 441, truncated = 0
+0.31.461.907 I srv  update_slots: all slots are idle
+0.31.524.318 I srv  params_from_: Chat format: peg-native
+0.31.526.203 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.921
+0.31.526.820 I reasoning-budget: activated, budget=2147483647 tokens
+0.31.526.906 I slot launch_slot_: id  0 | task 202 | processing task, is_child = 0
+0.31.526.929 W slot update_slots: id  0 | task 202 | n_past = 406, slot.prompt.tokens.size() = 441, seq_id = 0, pos_min = 440, n_swa = 0
+0.31.526.933 I slot update_slots: id  0 | task 202 | Checking checkpoint with [420, 420] against 406...
+0.31.526.935 W slot update_slots: id  0 | task 202 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.31.526.940 W slot update_slots: id  0 | task 202 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.32.129.825 I slot create_check: id  0 | task 202 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.32.519.469 I reasoning-budget: deactivated (natural end)
+0.33.361.317 I slot print_timing: id  0 | task 202 | 
+prompt eval time =     645.36 ms /   422 tokens (    1.53 ms per token,   653.90 tokens per second)
+       eval time =    1189.02 ms /    55 tokens (   21.62 ms per token,    46.26 tokens per second)
+      total time =    1834.38 ms /   477 tokens
+0.33.361.399 I slot      release: id  0 | task 202 | stop processing: n_tokens = 476, truncated = 0
+0.33.361.431 I srv  update_slots: all slots are idle
+0.33.406.059 I srv  params_from_: Chat format: peg-native
+0.33.406.536 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.854 (> 0.100 thold), f_keep = 0.884
+0.33.406.842 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.406.887 I slot launch_slot_: id  0 | task 259 | processing task, is_child = 0
+0.33.406.903 W slot update_slots: id  0 | task 259 | n_past = 421, slot.prompt.tokens.size() = 476, seq_id = 0, pos_min = 475, n_swa = 0
+0.33.406.904 I slot update_slots: id  0 | task 259 | Checking checkpoint with [417, 417] against 421...
+0.33.410.865 W slot update_slots: id  0 | task 259 | restored context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_past = 418, size = 62.813 MiB)
+0.33.668.081 I slot create_check: id  0 | task 259 | created context checkpoint 2 of 32 (pos_min = 488, pos_max = 488, n_tokens = 489, size = 62.813 MiB)
+0.34.301.983 I reasoning-budget: deactivated (natural end)
+0.34.590.965 I slot print_timing: id  0 | task 259 | 
+prompt eval time =     311.59 ms /    75 tokens (    4.15 ms per token,   240.70 tokens per second)
+       eval time =     872.45 ms /    38 tokens (   22.96 ms per token,    43.56 tokens per second)
+      total time =    1184.04 ms /   113 tokens
+0.34.591.058 I slot      release: id  0 | task 259 | stop processing: n_tokens = 530, truncated = 0
+0.34.591.093 I srv  update_slots: all slots are idle
+0.34.626.053 I srv  params_from_: Chat format: peg-native
+0.34.626.459 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.972 (> 0.100 thold), f_keep = 0.774
+0.34.626.655 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.626.694 I slot launch_slot_: id  0 | task 299 | processing task, is_child = 0
+0.34.626.705 W slot update_slots: id  0 | task 299 | n_past = 410, slot.prompt.tokens.size() = 530, seq_id = 0, pos_min = 529, n_swa = 0
+0.34.626.705 I slot update_slots: id  0 | task 299 | Checking checkpoint with [488, 488] against 410...
+0.34.626.706 I slot update_slots: id  0 | task 299 | Checking checkpoint with [417, 417] against 410...
+0.34.626.706 W slot update_slots: id  0 | task 299 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.34.626.709 W slot update_slots: id  0 | task 299 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.34.627.674 W slot update_slots: id  0 | task 299 | erased invalidated context checkpoint (pos_min = 488, pos_max = 488, n_tokens = 489, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.160.501 I slot create_check: id  0 | task 299 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.35.462.626 I reasoning-budget: deactivated (natural end)
+0.36.334.370 I slot print_timing: id  0 | task 299 | 
+prompt eval time =     570.77 ms /   422 tokens (    1.35 ms per token,   739.35 tokens per second)
+       eval time =    1136.85 ms /    53 tokens (   21.45 ms per token,    46.62 tokens per second)
+      total time =    1707.63 ms /   475 tokens
+0.36.334.535 I slot      release: id  0 | task 299 | stop processing: n_tokens = 474, truncated = 0
+0.36.334.587 I srv  update_slots: all slots are idle
+0.36.346.401 I srv  params_from_: Chat format: peg-native
+0.36.346.871 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 0.854
+0.36.347.275 I reasoning-budget: activated, budget=2147483647 tokens
+0.36.347.355 I slot launch_slot_: id  0 | task 354 | processing task, is_child = 0
+0.36.347.371 W slot update_slots: id  0 | task 354 | n_past = 405, slot.prompt.tokens.size() = 474, seq_id = 0, pos_min = 473, n_swa = 0
+0.36.347.374 I slot update_slots: id  0 | task 354 | Checking checkpoint with [417, 417] against 405...
+0.36.347.376 W slot update_slots: id  0 | task 354 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.36.347.381 W slot update_slots: id  0 | task 354 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.950.505 I slot create_check: id  0 | task 354 | created context checkpoint 1 of 32 (pos_min = 428, pos_max = 428, n_tokens = 429, size = 62.813 MiB)
+0.37.901.136 I reasoning-budget: deactivated (natural end)
+0.39.133.735 I slot print_timing: id  0 | task 354 | n_decoded =    100, tg =  46.90 t/s
+0.39.607.166 I slot print_timing: id  0 | task 354 | 
+prompt eval time =     653.96 ms /   433 tokens (    1.51 ms per token,   662.12 tokens per second)
+       eval time =    2605.80 ms /   123 tokens (   21.19 ms per token,    47.20 tokens per second)
+      total time =    3259.76 ms /   556 tokens
+0.39.607.372 I slot      release: id  0 | task 354 | stop processing: n_tokens = 555, truncated = 0
+0.39.607.435 I srv  update_slots: all slots are idle
+0.39.655.799 I srv  params_from_: Chat format: peg-native
+0.39.656.354 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.730
+0.39.656.713 I reasoning-budget: activated, budget=2147483647 tokens
+0.39.656.717 I reasoning-budget: deactivated (natural end)
+0.39.656.803 I slot launch_slot_: id  0 | task 479 | processing task, is_child = 0
+0.39.656.826 W slot update_slots: id  0 | task 479 | n_past = 405, slot.prompt.tokens.size() = 555, seq_id = 0, pos_min = 554, n_swa = 0
+0.39.656.828 I slot update_slots: id  0 | task 479 | Checking checkpoint with [428, 428] against 405...
+0.39.656.830 W slot update_slots: id  0 | task 479 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.39.656.835 W slot update_slots: id  0 | task 479 | erased invalidated context checkpoint (pos_min = 428, pos_max = 428, n_tokens = 429, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.40.205.076 I slot create_check: id  0 | task 479 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.41.065.168 I slot print_timing: id  0 | task 479 | 
+prompt eval time =     582.67 ms /   424 tokens (    1.37 ms per token,   727.68 tokens per second)
+       eval time =     825.66 ms /    39 tokens (   21.17 ms per token,    47.23 tokens per second)
+      total time =    1408.33 ms /   463 tokens
+0.41.065.250 I slot      release: id  0 | task 479 | stop processing: n_tokens = 462, truncated = 0
+0.41.065.277 I srv  update_slots: all slots are idle
+0.41.077.853 I srv  params_from_: Chat format: peg-native
+0.41.078.228 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.902 (> 0.100 thold), f_keep = 0.877
+0.41.078.482 I reasoning-budget: activated, budget=2147483647 tokens
+0.41.078.484 I reasoning-budget: deactivated (natural end)
+0.41.078.524 I slot launch_slot_: id  0 | task 520 | processing task, is_child = 0
+0.41.078.534 W slot update_slots: id  0 | task 520 | n_past = 405, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.41.078.535 I slot update_slots: id  0 | task 520 | Checking checkpoint with [419, 419] against 405...
+0.41.078.536 W slot update_slots: id  0 | task 520 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.41.078.537 W slot update_slots: id  0 | task 520 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.41.589.485 I slot create_check: id  0 | task 520 | created context checkpoint 1 of 32 (pos_min = 444, pos_max = 444, n_tokens = 445, size = 62.813 MiB)
+0.43.355.441 I slot print_timing: id  0 | task 520 | 
+prompt eval time =     549.58 ms /   449 tokens (    1.22 ms per token,   816.99 tokens per second)
+       eval time =    1727.31 ms /    86 tokens (   20.09 ms per token,    49.79 tokens per second)
+      total time =    2276.89 ms /   535 tokens
+0.43.355.528 I slot      release: id  0 | task 520 | stop processing: n_tokens = 534, truncated = 0
+0.43.355.559 I srv  update_slots: all slots are idle
+0.43.407.484 I srv  params_from_: Chat format: peg-native
+0.43.408.099 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.758
+0.43.408.372 I reasoning-budget: activated, budget=2147483647 tokens
+0.43.408.374 I reasoning-budget: deactivated (natural end)
+0.43.408.422 I slot launch_slot_: id  0 | task 608 | processing task, is_child = 0
+0.43.408.434 W slot update_slots: id  0 | task 608 | n_past = 405, slot.prompt.tokens.size() = 534, seq_id = 0, pos_min = 533, n_swa = 0
+0.43.408.435 I slot update_slots: id  0 | task 608 | Checking checkpoint with [444, 444] against 405...
+0.43.408.437 W slot update_slots: id  0 | task 608 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.43.408.440 W slot update_slots: id  0 | task 608 | erased invalidated context checkpoint (pos_min = 444, pos_max = 444, n_tokens = 445, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.43.970.219 I slot create_check: id  0 | task 608 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.44.791.545 I slot print_timing: id  0 | task 608 | 
+prompt eval time =     601.11 ms /   427 tokens (    1.41 ms per token,   710.35 tokens per second)
+       eval time =     781.99 ms /    39 tokens (   20.05 ms per token,    49.87 tokens per second)
+      total time =    1383.10 ms /   466 tokens
+0.44.791.624 I slot      release: id  0 | task 608 | stop processing: n_tokens = 465, truncated = 0
+0.44.791.654 I srv  update_slots: all slots are idle
+0.44.833.558 I srv  params_from_: Chat format: peg-native
+0.44.834.049 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.871
+0.44.834.271 I reasoning-budget: activated, budget=2147483647 tokens
+0.44.834.272 I reasoning-budget: deactivated (natural end)
+0.44.834.315 I slot launch_slot_: id  0 | task 649 | processing task, is_child = 0
+0.44.834.326 W slot update_slots: id  0 | task 649 | n_past = 405, slot.prompt.tokens.size() = 465, seq_id = 0, pos_min = 464, n_swa = 0
+0.44.834.327 I slot update_slots: id  0 | task 649 | Checking checkpoint with [422, 422] against 405...
+0.44.834.328 W slot update_slots: id  0 | task 649 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.44.834.332 W slot update_slots: id  0 | task 649 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.45.316.040 I slot create_check: id  0 | task 649 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.45.454.550 I slot print_timing: id  0 | task 649 | 
+prompt eval time =     522.74 ms /   427 tokens (    1.22 ms per token,   816.85 tokens per second)
+       eval time =      97.47 ms /     4 tokens (   24.37 ms per token,    41.04 tokens per second)
+      total time =     620.21 ms /   431 tokens
+0.45.454.641 I slot      release: id  0 | task 649 | stop processing: n_tokens = 430, truncated = 0
+0.45.454.670 I srv  update_slots: all slots are idle
+0.45.491.620 I srv  params_from_: Chat format: peg-native
+0.45.492.052 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.944
+0.45.492.301 I reasoning-budget: activated, budget=2147483647 tokens
+0.45.492.303 I reasoning-budget: deactivated (natural end)
+0.45.492.343 I slot launch_slot_: id  0 | task 655 | processing task, is_child = 0
+0.45.492.355 W slot update_slots: id  0 | task 655 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.45.492.357 I slot update_slots: id  0 | task 655 | Checking checkpoint with [422, 422] against 406...
+0.45.492.358 W slot update_slots: id  0 | task 655 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.45.492.362 W slot update_slots: id  0 | task 655 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.46.038.260 I slot create_check: id  0 | task 655 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.46.950.011 I slot print_timing: id  0 | task 655 | 
+prompt eval time =     592.07 ms /   424 tokens (    1.40 ms per token,   716.14 tokens per second)
+       eval time =     865.56 ms /    40 tokens (   21.64 ms per token,    46.21 tokens per second)
+      total time =    1457.62 ms /   464 tokens
+0.46.950.215 I slot      release: id  0 | task 655 | stop processing: n_tokens = 463, truncated = 0
+0.46.950.270 I srv  update_slots: all slots are idle
+0.46.968.851 I srv  params_from_: Chat format: peg-native
+0.46.969.311 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 1.000
+0.46.969.946 I reasoning-budget: activated, budget=2147483647 tokens
+0.46.969.951 I reasoning-budget: deactivated (natural end)
+0.46.970.047 I slot launch_slot_: id  0 | task 697 | processing task, is_child = 0
+0.47.121.064 I slot create_check: id  0 | task 697 | created context checkpoint 2 of 32 (pos_min = 490, pos_max = 490, n_tokens = 491, size = 62.813 MiB)
+0.47.570.159 I slot print_timing: id  0 | task 697 | 
+prompt eval time =     211.63 ms /    32 tokens (    6.61 ms per token,   151.21 tokens per second)
+       eval time =     388.43 ms /    15 tokens (   25.90 ms per token,    38.62 tokens per second)
+      total time =     600.06 ms /    47 tokens
+0.47.570.388 I slot      release: id  0 | task 697 | stop processing: n_tokens = 509, truncated = 0
+0.47.570.448 I srv  update_slots: all slots are idle
+0.47.607.526 I srv  params_from_: Chat format: peg-native
+0.47.608.040 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.806
+0.47.608.345 I reasoning-budget: activated, budget=2147483647 tokens
+0.47.608.347 I reasoning-budget: deactivated (natural end)
+0.47.608.398 I slot launch_slot_: id  0 | task 714 | processing task, is_child = 0
+0.47.608.410 W slot update_slots: id  0 | task 714 | n_past = 410, slot.prompt.tokens.size() = 509, seq_id = 0, pos_min = 508, n_swa = 0
+0.47.608.412 I slot update_slots: id  0 | task 714 | Checking checkpoint with [490, 490] against 410...
+0.47.608.412 I slot update_slots: id  0 | task 714 | Checking checkpoint with [419, 419] against 410...
+0.47.608.413 W slot update_slots: id  0 | task 714 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.47.608.416 W slot update_slots: id  0 | task 714 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.47.609.303 W slot update_slots: id  0 | task 714 | erased invalidated context checkpoint (pos_min = 490, pos_max = 490, n_tokens = 491, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.48.271.002 I slot create_check: id  0 | task 714 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.49.108.254 I slot print_timing: id  0 | task 714 | 
+prompt eval time =     731.88 ms /   424 tokens (    1.73 ms per token,   579.33 tokens per second)
+       eval time =     767.92 ms /    40 tokens (   19.20 ms per token,    52.09 tokens per second)
+      total time =    1499.80 ms /   464 tokens
+0.49.108.437 I slot      release: id  0 | task 714 | stop processing: n_tokens = 463, truncated = 0
+0.49.108.493 I srv  update_slots: all slots are idle
+0.49.161.163 I srv  params_from_: Chat format: peg-native
+0.49.163.186 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.931 (> 0.100 thold), f_keep = 0.875
+0.49.163.878 I reasoning-budget: activated, budget=2147483647 tokens
+0.49.163.884 I reasoning-budget: deactivated (natural end)
+0.49.163.990 I slot launch_slot_: id  0 | task 756 | processing task, is_child = 0
+0.49.164.016 W slot update_slots: id  0 | task 756 | n_past = 405, slot.prompt.tokens.size() = 463, seq_id = 0, pos_min = 462, n_swa = 0
+0.49.164.018 I slot update_slots: id  0 | task 756 | Checking checkpoint with [419, 419] against 405...
+0.49.164.020 W slot update_slots: id  0 | task 756 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.49.164.027 W slot update_slots: id  0 | task 756 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.49.744.671 I slot create_check: id  0 | task 756 | created context checkpoint 1 of 32 (pos_min = 430, pos_max = 430, n_tokens = 431, size = 62.813 MiB)
+0.51.381.802 I slot print_timing: id  0 | task 756 | 
+prompt eval time =     632.00 ms /   435 tokens (    1.45 ms per token,   688.30 tokens per second)
+       eval time =    1585.77 ms /    80 tokens (   19.82 ms per token,    50.45 tokens per second)
+      total time =    2217.77 ms /   515 tokens
+0.51.382.017 I slot      release: id  0 | task 756 | stop processing: n_tokens = 514, truncated = 0
+0.51.382.050 I srv  update_slots: all slots are idle
+0.51.383.517 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-roff-r3.log

@@ -0,0 +1,302 @@
+0.00.114.439 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.114.444 I device_info:
+0.00.114.555 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 123723 MiB free)
+0.00.114.708 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.114.715 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.114.796 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.114.847 I srv          init: running without SSL
+0.00.114.871 I srv          init: using 31 threads for HTTP server
+0.00.114.873 I srv          init: the WebUI is disabled
+0.00.114.942 I srv         start: binding port with default address family
+0.00.116.175 I srv          main: loading model
+0.00.116.182 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.160.730 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.21.513.834 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.793.651 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.217.020 I srv    load_model: initializing slots, n_slots = 1
+0.22.487.377 W srv    load_model: speculative decoding will use checkpoints
+0.22.487.387 W common_speculative_init: no implementations specified for speculative decoding
+0.22.487.390 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.22.487.471 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.22.487.473 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.22.487.495 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.22.540.918 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+'
+0.22.582.904 I srv          init: init: chat template, thinking = 0
+0.22.582.984 I srv          main: model loaded
+0.22.582.995 I srv          main: server is listening on http://127.0.0.1:18600
+0.22.583.003 I srv  update_slots: all slots are idle
+0.23.994.344 I srv  params_from_: Chat format: peg-native
+0.23.995.890 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.23.995.893 I srv  get_availabl: updating prompt cache
+0.23.995.900 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.23.995.907 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.23.995.909 I srv  get_availabl: prompt cache update took 0.01 ms
+0.23.996.222 I reasoning-budget: activated, budget=2147483647 tokens
+0.23.996.243 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.656.206 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.25.028.892 I reasoning-budget: deactivated (natural end)
+0.25.816.313 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     698.88 ms /   422 tokens (    1.66 ms per token,   603.83 tokens per second)
+       eval time =    1121.16 ms /    54 tokens (   20.76 ms per token,    48.16 tokens per second)
+      total time =    1820.04 ms /   476 tokens
+0.25.816.410 I slot      release: id  0 | task 0 | stop processing: n_tokens = 475, truncated = 0
+0.25.816.424 I srv  update_slots: all slots are idle
+0.25.830.282 I srv  params_from_: Chat format: peg-native
+0.25.830.661 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.906 (> 0.100 thold), f_keep = 0.853
+0.25.830.794 I reasoning-budget: activated, budget=2147483647 tokens
+0.25.830.832 I slot launch_slot_: id  0 | task 56 | processing task, is_child = 0
+0.25.830.841 W slot update_slots: id  0 | task 56 | n_past = 405, slot.prompt.tokens.size() = 475, seq_id = 0, pos_min = 474, n_swa = 0
+0.25.830.842 I slot update_slots: id  0 | task 56 | Checking checkpoint with [417, 417] against 405...
+0.25.830.844 W slot update_slots: id  0 | task 56 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.25.830.846 W slot update_slots: id  0 | task 56 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.26.482.806 I slot create_check: id  0 | task 56 | created context checkpoint 1 of 32 (pos_min = 442, pos_max = 442, n_tokens = 443, size = 62.813 MiB)
+0.26.903.330 I reasoning-budget: deactivated (natural end)
+0.28.669.262 I slot print_timing: id  0 | task 56 | n_decoded =    100, tg =  47.24 t/s
+0.28.729.150 I slot print_timing: id  0 | task 56 | 
+prompt eval time =     721.52 ms /   447 tokens (    1.61 ms per token,   619.52 tokens per second)
+       eval time =    2176.78 ms /   103 tokens (   21.13 ms per token,    47.32 tokens per second)
+      total time =    2898.30 ms /   550 tokens
+0.28.729.226 I slot      release: id  0 | task 56 | stop processing: n_tokens = 549, truncated = 0
+0.28.729.266 I srv  update_slots: all slots are idle
+0.28.742.730 I srv  params_from_: Chat format: peg-native
+0.28.743.168 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.738
+0.28.743.356 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.743.392 I slot launch_slot_: id  0 | task 161 | processing task, is_child = 0
+0.28.743.406 W slot update_slots: id  0 | task 161 | n_past = 405, slot.prompt.tokens.size() = 549, seq_id = 0, pos_min = 548, n_swa = 0
+0.28.743.407 I slot update_slots: id  0 | task 161 | Checking checkpoint with [442, 442] against 405...
+0.28.743.407 W slot update_slots: id  0 | task 161 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.743.411 W slot update_slots: id  0 | task 161 | erased invalidated context checkpoint (pos_min = 442, pos_max = 442, n_tokens = 443, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.29.274.427 I slot create_check: id  0 | task 161 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.29.681.788 I reasoning-budget: deactivated (natural end)
+0.30.465.485 I slot print_timing: id  0 | task 161 | 
+prompt eval time =     568.14 ms /   425 tokens (    1.34 ms per token,   748.05 tokens per second)
+       eval time =    1153.92 ms /    57 tokens (   20.24 ms per token,    49.40 tokens per second)
+      total time =    1722.06 ms /   482 tokens
+0.30.465.571 I slot      release: id  0 | task 161 | stop processing: n_tokens = 481, truncated = 0
+0.30.465.599 I srv  update_slots: all slots are idle
+0.30.480.783 I srv  params_from_: Chat format: peg-native
+0.30.481.142 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.842
+0.30.481.346 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.481.386 I slot launch_slot_: id  0 | task 220 | processing task, is_child = 0
+0.30.481.397 W slot update_slots: id  0 | task 220 | n_past = 405, slot.prompt.tokens.size() = 481, seq_id = 0, pos_min = 480, n_swa = 0
+0.30.481.397 I slot update_slots: id  0 | task 220 | Checking checkpoint with [420, 420] against 405...
+0.30.481.399 W slot update_slots: id  0 | task 220 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.481.401 W slot update_slots: id  0 | task 220 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.31.008.867 I slot create_check: id  0 | task 220 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.31.355.463 I reasoning-budget: deactivated (natural end)
+0.31.458.677 I slot print_timing: id  0 | task 220 | 
+prompt eval time =     568.77 ms /   425 tokens (    1.34 ms per token,   747.23 tokens per second)
+       eval time =     408.50 ms /    20 tokens (   20.42 ms per token,    48.96 tokens per second)
+      total time =     977.27 ms /   445 tokens
+0.31.458.774 I slot      release: id  0 | task 220 | stop processing: n_tokens = 444, truncated = 0
+0.31.458.805 I srv  update_slots: all slots are idle
+0.31.510.659 I srv  params_from_: Chat format: peg-native
+0.31.511.238 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.914
+0.31.511.571 I reasoning-budget: activated, budget=2147483647 tokens
+0.31.511.663 I slot launch_slot_: id  0 | task 242 | processing task, is_child = 0
+0.31.511.677 W slot update_slots: id  0 | task 242 | n_past = 406, slot.prompt.tokens.size() = 444, seq_id = 0, pos_min = 443, n_swa = 0
+0.31.511.684 I slot update_slots: id  0 | task 242 | Checking checkpoint with [420, 420] against 406...
+0.31.511.685 W slot update_slots: id  0 | task 242 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.31.511.690 W slot update_slots: id  0 | task 242 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.32.149.342 I slot create_check: id  0 | task 242 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.32.405.933 I reasoning-budget: deactivated (natural end)
+0.33.283.206 I slot print_timing: id  0 | task 242 | 
+prompt eval time =     677.52 ms /   422 tokens (    1.61 ms per token,   622.86 tokens per second)
+       eval time =    1093.99 ms /    52 tokens (   21.04 ms per token,    47.53 tokens per second)
+      total time =    1771.51 ms /   474 tokens
+0.33.283.283 I slot      release: id  0 | task 242 | stop processing: n_tokens = 473, truncated = 0
+0.33.283.309 I srv  update_slots: all slots are idle
+0.33.311.488 I srv  params_from_: Chat format: peg-native
+0.33.312.015 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.854 (> 0.100 thold), f_keep = 0.890
+0.33.312.441 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.312.556 I slot launch_slot_: id  0 | task 296 | processing task, is_child = 0
+0.33.312.580 W slot update_slots: id  0 | task 296 | n_past = 421, slot.prompt.tokens.size() = 473, seq_id = 0, pos_min = 472, n_swa = 0
+0.33.312.583 I slot update_slots: id  0 | task 296 | Checking checkpoint with [417, 417] against 421...
+0.33.321.205 W slot update_slots: id  0 | task 296 | restored context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_past = 418, size = 62.813 MiB)
+0.33.558.612 I slot create_check: id  0 | task 296 | created context checkpoint 2 of 32 (pos_min = 488, pos_max = 488, n_tokens = 489, size = 62.813 MiB)
+0.34.218.104 I reasoning-budget: deactivated (natural end)
+0.34.556.664 I slot print_timing: id  0 | task 296 | 
+prompt eval time =     286.32 ms /    75 tokens (    3.82 ms per token,   261.95 tokens per second)
+       eval time =     957.75 ms /    41 tokens (   23.36 ms per token,    42.81 tokens per second)
+      total time =    1244.07 ms /   116 tokens
+0.34.556.763 I slot      release: id  0 | task 296 | stop processing: n_tokens = 533, truncated = 0
+0.34.556.805 I srv  update_slots: all slots are idle
+0.34.571.001 I srv  params_from_: Chat format: peg-native
+0.34.571.465 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.972 (> 0.100 thold), f_keep = 0.769
+0.34.571.696 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.571.740 I slot launch_slot_: id  0 | task 339 | processing task, is_child = 0
+0.34.571.755 W slot update_slots: id  0 | task 339 | n_past = 410, slot.prompt.tokens.size() = 533, seq_id = 0, pos_min = 532, n_swa = 0
+0.34.571.756 I slot update_slots: id  0 | task 339 | Checking checkpoint with [488, 488] against 410...
+0.34.571.758 I slot update_slots: id  0 | task 339 | Checking checkpoint with [417, 417] against 410...
+0.34.571.759 W slot update_slots: id  0 | task 339 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.34.571.763 W slot update_slots: id  0 | task 339 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.34.573.311 W slot update_slots: id  0 | task 339 | erased invalidated context checkpoint (pos_min = 488, pos_max = 488, n_tokens = 489, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.110.746 I slot create_check: id  0 | task 339 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.35.584.142 I reasoning-budget: deactivated (natural end)
+0.36.392.408 I slot print_timing: id  0 | task 339 | 
+prompt eval time =     576.37 ms /   422 tokens (    1.37 ms per token,   732.16 tokens per second)
+       eval time =    1244.26 ms /    62 tokens (   20.07 ms per token,    49.83 tokens per second)
+      total time =    1820.64 ms /   484 tokens
+0.36.392.474 I slot      release: id  0 | task 339 | stop processing: n_tokens = 483, truncated = 0
+0.36.392.500 I srv  update_slots: all slots are idle
+0.36.404.680 I srv  params_from_: Chat format: peg-native
+0.36.405.094 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 0.839
+0.36.405.518 I reasoning-budget: activated, budget=2147483647 tokens
+0.36.405.599 I slot launch_slot_: id  0 | task 403 | processing task, is_child = 0
+0.36.405.620 W slot update_slots: id  0 | task 403 | n_past = 405, slot.prompt.tokens.size() = 483, seq_id = 0, pos_min = 482, n_swa = 0
+0.36.405.621 I slot update_slots: id  0 | task 403 | Checking checkpoint with [417, 417] against 405...
+0.36.405.625 W slot update_slots: id  0 | task 403 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.36.405.630 W slot update_slots: id  0 | task 403 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.974.524 I slot create_check: id  0 | task 403 | created context checkpoint 1 of 32 (pos_min = 428, pos_max = 428, n_tokens = 429, size = 62.813 MiB)
+0.39.070.997 I slot print_timing: id  0 | task 403 | n_decoded =    100, tg =  48.88 t/s
+0.39.448.467 I reasoning-budget: deactivated (natural end)
+0.41.042.559 I slot print_timing: id  0 | task 403 | 
+prompt eval time =     619.66 ms /   433 tokens (    1.43 ms per token,   698.77 tokens per second)
+       eval time =    4017.27 ms /   200 tokens (   20.09 ms per token,    49.79 tokens per second)
+      total time =    4636.93 ms /   633 tokens
+0.41.042.634 I slot      release: id  0 | task 403 | stop processing: n_tokens = 632, truncated = 0
+0.41.042.665 I srv  update_slots: all slots are idle
+0.41.054.733 I srv  params_from_: Chat format: peg-native
+0.41.055.120 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.641
+0.41.055.469 I reasoning-budget: activated, budget=2147483647 tokens
+0.41.055.471 I reasoning-budget: deactivated (natural end)
+0.41.055.514 I slot launch_slot_: id  0 | task 605 | processing task, is_child = 0
+0.41.055.527 W slot update_slots: id  0 | task 605 | n_past = 405, slot.prompt.tokens.size() = 632, seq_id = 0, pos_min = 631, n_swa = 0
+0.41.055.529 I slot update_slots: id  0 | task 605 | Checking checkpoint with [428, 428] against 405...
+0.41.055.530 W slot update_slots: id  0 | task 605 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.41.055.533 W slot update_slots: id  0 | task 605 | erased invalidated context checkpoint (pos_min = 428, pos_max = 428, n_tokens = 429, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.41.597.061 I slot create_check: id  0 | task 605 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.42.459.956 I slot print_timing: id  0 | task 605 | 
+prompt eval time =     582.62 ms /   424 tokens (    1.37 ms per token,   727.74 tokens per second)
+       eval time =     821.78 ms /    39 tokens (   21.07 ms per token,    47.46 tokens per second)
+      total time =    1404.41 ms /   463 tokens
+0.42.460.047 I slot      release: id  0 | task 605 | stop processing: n_tokens = 462, truncated = 0
+0.42.460.077 I srv  update_slots: all slots are idle
+0.42.475.720 I srv  params_from_: Chat format: peg-native
+0.42.476.233 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.902 (> 0.100 thold), f_keep = 0.877
+0.42.476.486 I reasoning-budget: activated, budget=2147483647 tokens
+0.42.476.491 I reasoning-budget: deactivated (natural end)
+0.42.476.536 I slot launch_slot_: id  0 | task 646 | processing task, is_child = 0
+0.42.476.550 W slot update_slots: id  0 | task 646 | n_past = 405, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.42.476.551 I slot update_slots: id  0 | task 646 | Checking checkpoint with [419, 419] against 405...
+0.42.476.553 W slot update_slots: id  0 | task 646 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.42.476.571 W slot update_slots: id  0 | task 646 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.43.126.158 I slot create_check: id  0 | task 646 | created context checkpoint 1 of 32 (pos_min = 444, pos_max = 444, n_tokens = 445, size = 62.813 MiB)
+0.44.988.511 I slot print_timing: id  0 | task 646 | 
+prompt eval time =     693.47 ms /   449 tokens (    1.54 ms per token,   647.47 tokens per second)
+       eval time =    1818.46 ms /    86 tokens (   21.14 ms per token,    47.29 tokens per second)
+      total time =    2511.93 ms /   535 tokens
+0.44.988.703 I slot      release: id  0 | task 646 | stop processing: n_tokens = 534, truncated = 0
+0.44.988.760 I srv  update_slots: all slots are idle
+0.45.002.229 I srv  params_from_: Chat format: peg-native
+0.45.002.631 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.758
+0.45.003.110 I reasoning-budget: activated, budget=2147483647 tokens
+0.45.003.114 I reasoning-budget: deactivated (natural end)
+0.45.003.190 I slot launch_slot_: id  0 | task 734 | processing task, is_child = 0
+0.45.003.216 W slot update_slots: id  0 | task 734 | n_past = 405, slot.prompt.tokens.size() = 534, seq_id = 0, pos_min = 533, n_swa = 0
+0.45.003.220 I slot update_slots: id  0 | task 734 | Checking checkpoint with [444, 444] against 405...
+0.45.003.221 W slot update_slots: id  0 | task 734 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.45.003.227 W slot update_slots: id  0 | task 734 | erased invalidated context checkpoint (pos_min = 444, pos_max = 444, n_tokens = 445, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.45.561.942 I slot create_check: id  0 | task 734 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.46.436.833 I slot print_timing: id  0 | task 734 | 
+prompt eval time =     595.84 ms /   427 tokens (    1.40 ms per token,   716.64 tokens per second)
+       eval time =     837.77 ms /    39 tokens (   21.48 ms per token,    46.55 tokens per second)
+      total time =    1433.60 ms /   466 tokens
+0.46.436.923 I slot      release: id  0 | task 734 | stop processing: n_tokens = 465, truncated = 0
+0.46.436.955 I srv  update_slots: all slots are idle
+0.46.450.626 I srv  params_from_: Chat format: peg-native
+0.46.451.043 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.871
+0.46.451.527 I reasoning-budget: activated, budget=2147483647 tokens
+0.46.451.532 I reasoning-budget: deactivated (natural end)
+0.46.451.617 I slot launch_slot_: id  0 | task 775 | processing task, is_child = 0
+0.46.451.637 W slot update_slots: id  0 | task 775 | n_past = 405, slot.prompt.tokens.size() = 465, seq_id = 0, pos_min = 464, n_swa = 0
+0.46.451.641 I slot update_slots: id  0 | task 775 | Checking checkpoint with [422, 422] against 405...
+0.46.451.643 W slot update_slots: id  0 | task 775 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.46.451.649 W slot update_slots: id  0 | task 775 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.46.982.640 I slot create_check: id  0 | task 775 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.47.142.299 I slot print_timing: id  0 | task 775 | 
+prompt eval time =     585.65 ms /   427 tokens (    1.37 ms per token,   729.10 tokens per second)
+       eval time =     104.99 ms /     4 tokens (   26.25 ms per token,    38.10 tokens per second)
+      total time =     690.65 ms /   431 tokens
+0.47.142.414 I slot      release: id  0 | task 775 | stop processing: n_tokens = 430, truncated = 0
+0.47.142.449 I srv  update_slots: all slots are idle
+0.47.193.526 I srv  params_from_: Chat format: peg-native
+0.47.195.446 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.944
+0.47.195.987 I reasoning-budget: activated, budget=2147483647 tokens
+0.47.195.992 I reasoning-budget: deactivated (natural end)
+0.47.196.101 I slot launch_slot_: id  0 | task 781 | processing task, is_child = 0
+0.47.196.122 W slot update_slots: id  0 | task 781 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.47.196.128 I slot update_slots: id  0 | task 781 | Checking checkpoint with [422, 422] against 406...
+0.47.196.129 W slot update_slots: id  0 | task 781 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.47.196.139 W slot update_slots: id  0 | task 781 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.47.774.566 I slot create_check: id  0 | task 781 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.48.760.957 I slot print_timing: id  0 | task 781 | 
+prompt eval time =     635.18 ms /   424 tokens (    1.50 ms per token,   667.53 tokens per second)
+       eval time =     929.64 ms /    40 tokens (   23.24 ms per token,    43.03 tokens per second)
+      total time =    1564.83 ms /   464 tokens
+0.48.761.052 I slot      release: id  0 | task 781 | stop processing: n_tokens = 463, truncated = 0
+0.48.761.080 I srv  update_slots: all slots are idle
+0.48.783.508 I srv  params_from_: Chat format: peg-native
+0.48.783.860 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 1.000
+0.48.784.082 I reasoning-budget: activated, budget=2147483647 tokens
+0.48.784.084 I reasoning-budget: deactivated (natural end)
+0.48.784.127 I slot launch_slot_: id  0 | task 823 | processing task, is_child = 0
+0.48.935.107 I slot create_check: id  0 | task 823 | created context checkpoint 2 of 32 (pos_min = 490, pos_max = 490, n_tokens = 491, size = 62.813 MiB)
+0.49.277.790 I slot print_timing: id  0 | task 823 | 
+prompt eval time =     217.60 ms /    32 tokens (    6.80 ms per token,   147.06 tokens per second)
+       eval time =     276.04 ms /    13 tokens (   21.23 ms per token,    47.09 tokens per second)
+      total time =     493.64 ms /    45 tokens
+0.49.277.883 I slot      release: id  0 | task 823 | stop processing: n_tokens = 507, truncated = 0
+0.49.277.910 I srv  update_slots: all slots are idle
+0.49.324.008 I srv  params_from_: Chat format: peg-native
+0.49.324.548 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.809
+0.49.324.947 I reasoning-budget: activated, budget=2147483647 tokens
+0.49.324.952 I reasoning-budget: deactivated (natural end)
+0.49.325.029 I slot launch_slot_: id  0 | task 838 | processing task, is_child = 0
+0.49.325.041 W slot update_slots: id  0 | task 838 | n_past = 410, slot.prompt.tokens.size() = 507, seq_id = 0, pos_min = 506, n_swa = 0
+0.49.325.042 I slot update_slots: id  0 | task 838 | Checking checkpoint with [490, 490] against 410...
+0.49.325.043 I slot update_slots: id  0 | task 838 | Checking checkpoint with [419, 419] against 410...
+0.49.325.048 W slot update_slots: id  0 | task 838 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.49.325.051 W slot update_slots: id  0 | task 838 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.49.328.874 W slot update_slots: id  0 | task 838 | erased invalidated context checkpoint (pos_min = 490, pos_max = 490, n_tokens = 491, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.49.907.729 I slot create_check: id  0 | task 838 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.50.729.586 I slot print_timing: id  0 | task 838 | 
+prompt eval time =     624.90 ms /   424 tokens (    1.47 ms per token,   678.50 tokens per second)
+       eval time =     779.62 ms /    40 tokens (   19.49 ms per token,    51.31 tokens per second)
+      total time =    1404.52 ms /   464 tokens
+0.50.729.654 I slot      release: id  0 | task 838 | stop processing: n_tokens = 463, truncated = 0
+0.50.729.686 I srv  update_slots: all slots are idle
+0.50.775.784 I srv  params_from_: Chat format: peg-native
+0.50.776.334 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.931 (> 0.100 thold), f_keep = 0.875
+0.50.776.559 I reasoning-budget: activated, budget=2147483647 tokens
+0.50.776.562 I reasoning-budget: deactivated (natural end)
+0.50.776.609 I slot launch_slot_: id  0 | task 880 | processing task, is_child = 0
+0.50.776.620 W slot update_slots: id  0 | task 880 | n_past = 405, slot.prompt.tokens.size() = 463, seq_id = 0, pos_min = 462, n_swa = 0
+0.50.776.622 I slot update_slots: id  0 | task 880 | Checking checkpoint with [419, 419] against 405...
+0.50.776.623 W slot update_slots: id  0 | task 880 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.50.776.627 W slot update_slots: id  0 | task 880 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.51.328.909 I slot create_check: id  0 | task 880 | created context checkpoint 1 of 32 (pos_min = 430, pos_max = 430, n_tokens = 431, size = 62.813 MiB)
+0.52.962.266 I slot print_timing: id  0 | task 880 | 
+prompt eval time =     596.05 ms /   435 tokens (    1.37 ms per token,   729.81 tokens per second)
+       eval time =    1589.57 ms /    80 tokens (   19.87 ms per token,    50.33 tokens per second)
+      total time =    2185.62 ms /   515 tokens
+0.52.962.487 I slot      release: id  0 | task 880 | stop processing: n_tokens = 514, truncated = 0
+0.52.962.523 I srv  update_slots: all slots are idle
+0.52.963.931 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-roff.log

@@ -0,0 +1,301 @@
+0.00.077.092 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.077.102 I device_info:
+0.00.077.242 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 123819 MiB free)
+0.00.077.530 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.077.538 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.077.648 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.077.722 I srv          init: running without SSL
+0.00.077.768 I srv          init: using 31 threads for HTTP server
+0.00.077.770 I srv          init: the WebUI is disabled
+0.00.077.863 I srv         start: binding port with default address family
+0.00.079.093 I srv          main: loading model
+0.00.079.096 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.120.315 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.21.810.861 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.22.004.505 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.307.874 I srv    load_model: initializing slots, n_slots = 1
+0.22.526.550 W srv    load_model: speculative decoding will use checkpoints
+0.22.526.576 W common_speculative_init: no implementations specified for speculative decoding
+0.22.526.584 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.22.526.740 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.22.526.745 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.22.526.855 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.22.576.014 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+'
+0.22.604.308 I srv          init: init: chat template, thinking = 0
+0.22.604.360 I srv          main: model loaded
+0.22.604.364 I srv          main: server is listening on http://127.0.0.1:18600
+0.22.604.370 I srv  update_slots: all slots are idle
+0.24.142.693 I srv  params_from_: Chat format: peg-native
+0.24.143.568 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.143.571 I srv  get_availabl: updating prompt cache
+0.24.143.578 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.143.583 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.143.585 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.143.943 I reasoning-budget: activated, budget=2147483647 tokens
+0.24.143.962 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.903.283 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.25.273.393 I reasoning-budget: deactivated (natural end)
+0.26.099.925 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     810.27 ms /   422 tokens (    1.92 ms per token,   520.81 tokens per second)
+       eval time =    1145.63 ms /    51 tokens (   22.46 ms per token,    44.52 tokens per second)
+      total time =    1955.91 ms /   473 tokens
+0.26.100.146 I slot      release: id  0 | task 0 | stop processing: n_tokens = 472, truncated = 0
+0.26.100.163 I srv  update_slots: all slots are idle
+0.26.148.514 I srv  params_from_: Chat format: peg-native
+0.26.149.069 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.906 (> 0.100 thold), f_keep = 0.858
+0.26.149.246 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.149.294 I slot launch_slot_: id  0 | task 53 | processing task, is_child = 0
+0.26.149.305 W slot update_slots: id  0 | task 53 | n_past = 405, slot.prompt.tokens.size() = 472, seq_id = 0, pos_min = 471, n_swa = 0
+0.26.149.306 I slot update_slots: id  0 | task 53 | Checking checkpoint with [417, 417] against 405...
+0.26.149.307 W slot update_slots: id  0 | task 53 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.26.149.310 W slot update_slots: id  0 | task 53 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.26.723.793 I slot create_check: id  0 | task 53 | created context checkpoint 1 of 32 (pos_min = 442, pos_max = 442, n_tokens = 443, size = 62.813 MiB)
+0.27.315.116 I reasoning-budget: deactivated (natural end)
+0.28.126.542 I slot print_timing: id  0 | task 53 | 
+prompt eval time =     626.21 ms /   447 tokens (    1.40 ms per token,   713.82 tokens per second)
+       eval time =    1350.95 ms /    68 tokens (   19.87 ms per token,    50.34 tokens per second)
+      total time =    1977.16 ms /   515 tokens
+0.28.126.747 I slot      release: id  0 | task 53 | stop processing: n_tokens = 514, truncated = 0
+0.28.126.781 I srv  update_slots: all slots are idle
+0.28.164.510 I srv  params_from_: Chat format: peg-native
+0.28.166.615 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.788
+0.28.167.130 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.167.205 I slot launch_slot_: id  0 | task 123 | processing task, is_child = 0
+0.28.167.226 W slot update_slots: id  0 | task 123 | n_past = 405, slot.prompt.tokens.size() = 514, seq_id = 0, pos_min = 513, n_swa = 0
+0.28.167.229 I slot update_slots: id  0 | task 123 | Checking checkpoint with [442, 442] against 405...
+0.28.167.232 W slot update_slots: id  0 | task 123 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.167.239 W slot update_slots: id  0 | task 123 | erased invalidated context checkpoint (pos_min = 442, pos_max = 442, n_tokens = 443, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.745.049 I slot create_check: id  0 | task 123 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.28.952.307 I reasoning-budget: deactivated (natural end)
+0.29.756.441 I slot print_timing: id  0 | task 123 | 
+prompt eval time =     613.73 ms /   425 tokens (    1.44 ms per token,   692.49 tokens per second)
+       eval time =     975.48 ms /    48 tokens (   20.32 ms per token,    49.21 tokens per second)
+      total time =    1589.20 ms /   473 tokens
+0.29.756.519 I slot      release: id  0 | task 123 | stop processing: n_tokens = 472, truncated = 0
+0.29.756.549 I srv  update_slots: all slots are idle
+0.29.771.836 I srv  params_from_: Chat format: peg-native
+0.29.772.265 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.858
+0.29.772.460 I reasoning-budget: activated, budget=2147483647 tokens
+0.29.772.502 I slot launch_slot_: id  0 | task 173 | processing task, is_child = 0
+0.29.772.513 W slot update_slots: id  0 | task 173 | n_past = 405, slot.prompt.tokens.size() = 472, seq_id = 0, pos_min = 471, n_swa = 0
+0.29.772.513 I slot update_slots: id  0 | task 173 | Checking checkpoint with [420, 420] against 405...
+0.29.772.514 W slot update_slots: id  0 | task 173 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.29.772.518 W slot update_slots: id  0 | task 173 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.386.059 I slot create_check: id  0 | task 173 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.30.672.836 I reasoning-budget: deactivated (natural end)
+0.30.764.932 I slot print_timing: id  0 | task 173 | 
+prompt eval time =     647.59 ms /   425 tokens (    1.52 ms per token,   656.28 tokens per second)
+       eval time =     344.81 ms /    17 tokens (   20.28 ms per token,    49.30 tokens per second)
+      total time =     992.40 ms /   442 tokens
+0.30.765.023 I slot      release: id  0 | task 173 | stop processing: n_tokens = 441, truncated = 0
+0.30.765.050 I srv  update_slots: all slots are idle
+0.30.804.618 I srv  params_from_: Chat format: peg-native
+0.30.806.300 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.921
+0.30.806.744 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.806.810 I slot launch_slot_: id  0 | task 192 | processing task, is_child = 0
+0.30.806.826 W slot update_slots: id  0 | task 192 | n_past = 406, slot.prompt.tokens.size() = 441, seq_id = 0, pos_min = 440, n_swa = 0
+0.30.806.829 I slot update_slots: id  0 | task 192 | Checking checkpoint with [420, 420] against 406...
+0.30.806.830 W slot update_slots: id  0 | task 192 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.806.835 W slot update_slots: id  0 | task 192 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.31.351.459 I slot create_check: id  0 | task 192 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.31.884.045 I reasoning-budget: deactivated (natural end)
+0.32.727.839 I slot print_timing: id  0 | task 192 | 
+prompt eval time =     584.38 ms /   422 tokens (    1.38 ms per token,   722.13 tokens per second)
+       eval time =    1336.62 ms /    62 tokens (   21.56 ms per token,    46.39 tokens per second)
+      total time =    1921.00 ms /   484 tokens
+0.32.727.928 I slot      release: id  0 | task 192 | stop processing: n_tokens = 483, truncated = 0
+0.32.727.959 I srv  update_slots: all slots are idle
+0.32.742.237 I srv  params_from_: Chat format: peg-native
+0.32.742.652 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.854 (> 0.100 thold), f_keep = 0.872
+0.32.743.158 I reasoning-budget: activated, budget=2147483647 tokens
+0.32.743.231 I slot launch_slot_: id  0 | task 256 | processing task, is_child = 0
+0.32.743.251 W slot update_slots: id  0 | task 256 | n_past = 421, slot.prompt.tokens.size() = 483, seq_id = 0, pos_min = 482, n_swa = 0
+0.32.743.254 I slot update_slots: id  0 | task 256 | Checking checkpoint with [417, 417] against 421...
+0.32.748.604 W slot update_slots: id  0 | task 256 | restored context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_past = 418, size = 62.813 MiB)
+0.32.958.098 I slot create_check: id  0 | task 256 | created context checkpoint 2 of 32 (pos_min = 488, pos_max = 488, n_tokens = 489, size = 62.813 MiB)
+0.33.556.876 I reasoning-budget: deactivated (natural end)
+0.33.886.761 I slot print_timing: id  0 | task 256 | 
+prompt eval time =     258.37 ms /    75 tokens (    3.44 ms per token,   290.28 tokens per second)
+       eval time =     885.13 ms /    42 tokens (   21.07 ms per token,    47.45 tokens per second)
+      total time =    1143.50 ms /   117 tokens
+0.33.886.848 I slot      release: id  0 | task 256 | stop processing: n_tokens = 534, truncated = 0
+0.33.886.880 I srv  update_slots: all slots are idle
+0.33.924.683 I srv  params_from_: Chat format: peg-native
+0.33.925.111 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.972 (> 0.100 thold), f_keep = 0.768
+0.33.925.551 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.925.629 I slot launch_slot_: id  0 | task 300 | processing task, is_child = 0
+0.33.925.649 W slot update_slots: id  0 | task 300 | n_past = 410, slot.prompt.tokens.size() = 534, seq_id = 0, pos_min = 533, n_swa = 0
+0.33.925.650 I slot update_slots: id  0 | task 300 | Checking checkpoint with [488, 488] against 410...
+0.33.925.652 I slot update_slots: id  0 | task 300 | Checking checkpoint with [417, 417] against 410...
+0.33.925.653 W slot update_slots: id  0 | task 300 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.33.925.658 W slot update_slots: id  0 | task 300 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.33.927.212 W slot update_slots: id  0 | task 300 | erased invalidated context checkpoint (pos_min = 488, pos_max = 488, n_tokens = 489, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.34.478.068 I slot create_check: id  0 | task 300 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.34.936.115 I reasoning-budget: deactivated (natural end)
+0.35.820.698 I slot print_timing: id  0 | task 300 | 
+prompt eval time =     590.50 ms /   422 tokens (    1.40 ms per token,   714.65 tokens per second)
+       eval time =    1304.53 ms /    61 tokens (   21.39 ms per token,    46.76 tokens per second)
+      total time =    1895.03 ms /   483 tokens
+0.35.820.771 I slot      release: id  0 | task 300 | stop processing: n_tokens = 482, truncated = 0
+0.35.820.802 I srv  update_slots: all slots are idle
+0.35.833.343 I srv  params_from_: Chat format: peg-native
+0.35.833.721 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 0.840
+0.35.833.997 I reasoning-budget: activated, budget=2147483647 tokens
+0.35.834.089 I slot launch_slot_: id  0 | task 363 | processing task, is_child = 0
+0.35.834.109 W slot update_slots: id  0 | task 363 | n_past = 405, slot.prompt.tokens.size() = 482, seq_id = 0, pos_min = 481, n_swa = 0
+0.35.834.109 I slot update_slots: id  0 | task 363 | Checking checkpoint with [417, 417] against 405...
+0.35.834.111 W slot update_slots: id  0 | task 363 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.35.834.118 W slot update_slots: id  0 | task 363 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.300.366 I slot create_check: id  0 | task 363 | created context checkpoint 1 of 32 (pos_min = 428, pos_max = 428, n_tokens = 429, size = 62.813 MiB)
+0.36.903.862 I reasoning-budget: deactivated (natural end)
+0.38.352.136 I slot print_timing: id  0 | task 363 | n_decoded =    100, tg =  49.65 t/s
+0.38.552.887 I slot print_timing: id  0 | task 363 | 
+prompt eval time =     504.10 ms /   433 tokens (    1.16 ms per token,   858.96 tokens per second)
+       eval time =    2214.64 ms /   109 tokens (   20.32 ms per token,    49.22 tokens per second)
+      total time =    2718.74 ms /   542 tokens
+0.38.553.137 I slot      release: id  0 | task 363 | stop processing: n_tokens = 541, truncated = 0
+0.38.553.221 I srv  update_slots: all slots are idle
+0.38.576.788 I srv  params_from_: Chat format: peg-native
+0.38.577.342 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.749
+0.38.577.906 I reasoning-budget: activated, budget=2147483647 tokens
+0.38.577.907 I reasoning-budget: deactivated (natural end)
+0.38.577.985 I slot launch_slot_: id  0 | task 474 | processing task, is_child = 0
+0.38.578.015 W slot update_slots: id  0 | task 474 | n_past = 405, slot.prompt.tokens.size() = 541, seq_id = 0, pos_min = 540, n_swa = 0
+0.38.578.016 I slot update_slots: id  0 | task 474 | Checking checkpoint with [428, 428] against 405...
+0.38.578.018 W slot update_slots: id  0 | task 474 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.38.578.022 W slot update_slots: id  0 | task 474 | erased invalidated context checkpoint (pos_min = 428, pos_max = 428, n_tokens = 429, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.39.129.005 I slot create_check: id  0 | task 474 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.40.014.664 I slot print_timing: id  0 | task 474 | 
+prompt eval time =     601.68 ms /   424 tokens (    1.42 ms per token,   704.69 tokens per second)
+       eval time =     834.95 ms /    39 tokens (   21.41 ms per token,    46.71 tokens per second)
+      total time =    1436.64 ms /   463 tokens
+0.40.014.755 I slot      release: id  0 | task 474 | stop processing: n_tokens = 462, truncated = 0
+0.40.014.785 I srv  update_slots: all slots are idle
+0.40.063.657 I srv  params_from_: Chat format: peg-native
+0.40.065.988 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.902 (> 0.100 thold), f_keep = 0.877
+0.40.066.550 I reasoning-budget: activated, budget=2147483647 tokens
+0.40.066.555 I reasoning-budget: deactivated (natural end)
+0.40.066.646 I slot launch_slot_: id  0 | task 515 | processing task, is_child = 0
+0.40.066.666 W slot update_slots: id  0 | task 515 | n_past = 405, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.40.066.671 I slot update_slots: id  0 | task 515 | Checking checkpoint with [419, 419] against 405...
+0.40.066.673 W slot update_slots: id  0 | task 515 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.40.066.679 W slot update_slots: id  0 | task 515 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.40.651.450 I slot create_check: id  0 | task 515 | created context checkpoint 1 of 32 (pos_min = 444, pos_max = 444, n_tokens = 445, size = 62.813 MiB)
+0.42.596.129 I slot print_timing: id  0 | task 515 | 
+prompt eval time =     621.59 ms /   449 tokens (    1.38 ms per token,   722.34 tokens per second)
+       eval time =    1907.84 ms /    86 tokens (   22.18 ms per token,    45.08 tokens per second)
+      total time =    2529.43 ms /   535 tokens
+0.42.596.351 I slot      release: id  0 | task 515 | stop processing: n_tokens = 534, truncated = 0
+0.42.596.411 I srv  update_slots: all slots are idle
+0.42.613.151 I srv  params_from_: Chat format: peg-native
+0.42.613.661 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.758
+0.42.613.938 I reasoning-budget: activated, budget=2147483647 tokens
+0.42.613.941 I reasoning-budget: deactivated (natural end)
+0.42.613.986 I slot launch_slot_: id  0 | task 603 | processing task, is_child = 0
+0.42.614.003 W slot update_slots: id  0 | task 603 | n_past = 405, slot.prompt.tokens.size() = 534, seq_id = 0, pos_min = 533, n_swa = 0
+0.42.614.004 I slot update_slots: id  0 | task 603 | Checking checkpoint with [444, 444] against 405...
+0.42.614.006 W slot update_slots: id  0 | task 603 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.42.614.008 W slot update_slots: id  0 | task 603 | erased invalidated context checkpoint (pos_min = 444, pos_max = 444, n_tokens = 445, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.43.137.655 I slot create_check: id  0 | task 603 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.44.068.310 I slot print_timing: id  0 | task 603 | 
+prompt eval time =     589.04 ms /   427 tokens (    1.38 ms per token,   724.90 tokens per second)
+       eval time =     865.23 ms /    39 tokens (   22.19 ms per token,    45.07 tokens per second)
+      total time =    1454.27 ms /   466 tokens
+0.44.068.519 I slot      release: id  0 | task 603 | stop processing: n_tokens = 465, truncated = 0
+0.44.068.586 I srv  update_slots: all slots are idle
+0.44.108.932 I srv  params_from_: Chat format: peg-native
+0.44.109.519 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.871
+0.44.109.808 I reasoning-budget: activated, budget=2147483647 tokens
+0.44.109.810 I reasoning-budget: deactivated (natural end)
+0.44.109.863 I slot launch_slot_: id  0 | task 644 | processing task, is_child = 0
+0.44.109.876 W slot update_slots: id  0 | task 644 | n_past = 405, slot.prompt.tokens.size() = 465, seq_id = 0, pos_min = 464, n_swa = 0
+0.44.109.878 I slot update_slots: id  0 | task 644 | Checking checkpoint with [422, 422] against 405...
+0.44.109.879 W slot update_slots: id  0 | task 644 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.44.109.881 W slot update_slots: id  0 | task 644 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.44.678.393 I slot create_check: id  0 | task 644 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.44.867.520 I slot print_timing: id  0 | task 644 | 
+prompt eval time =     621.79 ms /   427 tokens (    1.46 ms per token,   686.72 tokens per second)
+       eval time =     135.82 ms /     4 tokens (   33.96 ms per token,    29.45 tokens per second)
+      total time =     757.61 ms /   431 tokens
+0.44.867.722 I slot      release: id  0 | task 644 | stop processing: n_tokens = 430, truncated = 0
+0.44.867.777 I srv  update_slots: all slots are idle
+0.44.920.558 I srv  params_from_: Chat format: peg-native
+0.44.922.823 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.944
+0.44.923.382 I reasoning-budget: activated, budget=2147483647 tokens
+0.44.923.386 I reasoning-budget: deactivated (natural end)
+0.44.923.473 I slot launch_slot_: id  0 | task 650 | processing task, is_child = 0
+0.44.923.490 W slot update_slots: id  0 | task 650 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.44.923.492 I slot update_slots: id  0 | task 650 | Checking checkpoint with [422, 422] against 406...
+0.44.923.496 W slot update_slots: id  0 | task 650 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.44.923.503 W slot update_slots: id  0 | task 650 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.45.494.241 I slot create_check: id  0 | task 650 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.46.360.864 I slot print_timing: id  0 | task 650 | 
+prompt eval time =     619.67 ms /   424 tokens (    1.46 ms per token,   684.24 tokens per second)
+       eval time =     817.69 ms /    40 tokens (   20.44 ms per token,    48.92 tokens per second)
+      total time =    1437.36 ms /   464 tokens
+0.46.360.962 I slot      release: id  0 | task 650 | stop processing: n_tokens = 463, truncated = 0
+0.46.360.994 I srv  update_slots: all slots are idle
+0.46.376.404 I srv  params_from_: Chat format: peg-native
+0.46.376.859 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 1.000
+0.46.377.382 I reasoning-budget: activated, budget=2147483647 tokens
+0.46.377.385 I reasoning-budget: deactivated (natural end)
+0.46.377.478 I slot launch_slot_: id  0 | task 692 | processing task, is_child = 0
+0.46.580.028 I slot create_check: id  0 | task 692 | created context checkpoint 2 of 32 (pos_min = 490, pos_max = 490, n_tokens = 491, size = 62.813 MiB)
+0.46.886.342 I slot print_timing: id  0 | task 692 | 
+prompt eval time =     260.43 ms /    32 tokens (    8.14 ms per token,   122.87 tokens per second)
+       eval time =     248.41 ms /    13 tokens (   19.11 ms per token,    52.33 tokens per second)
+      total time =     508.84 ms /    45 tokens
+0.46.886.415 I slot      release: id  0 | task 692 | stop processing: n_tokens = 507, truncated = 0
+0.46.886.437 I srv  update_slots: all slots are idle
+0.46.902.435 I srv  params_from_: Chat format: peg-native
+0.46.903.190 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.809
+0.46.903.420 I reasoning-budget: activated, budget=2147483647 tokens
+0.46.903.422 I reasoning-budget: deactivated (natural end)
+0.46.903.471 I slot launch_slot_: id  0 | task 707 | processing task, is_child = 0
+0.46.903.481 W slot update_slots: id  0 | task 707 | n_past = 410, slot.prompt.tokens.size() = 507, seq_id = 0, pos_min = 506, n_swa = 0
+0.46.903.483 I slot update_slots: id  0 | task 707 | Checking checkpoint with [490, 490] against 410...
+0.46.903.484 I slot update_slots: id  0 | task 707 | Checking checkpoint with [419, 419] against 410...
+0.46.903.485 W slot update_slots: id  0 | task 707 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.46.903.488 W slot update_slots: id  0 | task 707 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.46.904.537 W slot update_slots: id  0 | task 707 | erased invalidated context checkpoint (pos_min = 490, pos_max = 490, n_tokens = 491, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.47.441.027 I slot create_check: id  0 | task 707 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.48.289.348 I slot print_timing: id  0 | task 707 | 
+prompt eval time =     577.77 ms /   424 tokens (    1.36 ms per token,   733.85 tokens per second)
+       eval time =     808.09 ms /    40 tokens (   20.20 ms per token,    49.50 tokens per second)
+      total time =    1385.86 ms /   464 tokens
+0.48.289.404 I slot      release: id  0 | task 707 | stop processing: n_tokens = 463, truncated = 0
+0.48.289.426 I srv  update_slots: all slots are idle
+0.48.302.317 I srv  params_from_: Chat format: peg-native
+0.48.302.639 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.931 (> 0.100 thold), f_keep = 0.875
+0.48.302.821 I reasoning-budget: activated, budget=2147483647 tokens
+0.48.302.822 I reasoning-budget: deactivated (natural end)
+0.48.302.871 I slot launch_slot_: id  0 | task 749 | processing task, is_child = 0
+0.48.302.880 W slot update_slots: id  0 | task 749 | n_past = 405, slot.prompt.tokens.size() = 463, seq_id = 0, pos_min = 462, n_swa = 0
+0.48.302.881 I slot update_slots: id  0 | task 749 | Checking checkpoint with [419, 419] against 405...
+0.48.302.882 W slot update_slots: id  0 | task 749 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.48.302.884 W slot update_slots: id  0 | task 749 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.48.829.428 I slot create_check: id  0 | task 749 | created context checkpoint 1 of 32 (pos_min = 430, pos_max = 430, n_tokens = 431, size = 62.813 MiB)
+0.50.509.107 I slot print_timing: id  0 | task 749 | 
+prompt eval time =     572.46 ms /   435 tokens (    1.32 ms per token,   759.87 tokens per second)
+       eval time =    1633.74 ms /    80 tokens (   20.42 ms per token,    48.97 tokens per second)
+      total time =    2206.20 ms /   515 tokens
+0.50.509.286 I slot      release: id  0 | task 749 | stop processing: n_tokens = 514, truncated = 0
+0.50.509.314 I srv  update_slots: all slots are idle
+0.50.510.178 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-tpl-medium-probe.log

@@ -0,0 +1,278 @@
+0.00.129.990 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.129.993 I device_info:
+0.00.130.113 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 122319 MiB free)
+0.00.130.278 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.130.286 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.130.376 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.130.453 I srv          init: running without SSL
+0.00.130.481 I srv          init: using 31 threads for HTTP server
+0.00.130.482 I srv          init: the WebUI is disabled
+0.00.130.551 I srv         start: binding port with default address family
+0.00.131.742 I srv          main: loading model
+0.00.131.749 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.190.887 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.23.864.154 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.24.195.940 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.24.556.150 I srv    load_model: initializing slots, n_slots = 1
+0.24.849.854 W srv    load_model: speculative decoding will use checkpoints
+0.24.849.866 W common_speculative_init: no implementations specified for speculative decoding
+0.24.849.870 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.24.849.977 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.24.849.981 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.24.850.006 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.24.907.565 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.24.952.737 I srv          init: init: chat template, thinking = 1
+0.24.952.792 I srv          main: model loaded
+0.24.952.798 I srv          main: server is listening on http://127.0.0.1:18652
+0.24.952.803 I srv  update_slots: all slots are idle
+0.26.377.211 I srv  params_from_: Chat format: peg-native
+0.26.379.299 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.26.379.305 I srv  get_availabl: updating prompt cache
+0.26.379.315 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.26.379.323 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.26.379.326 I srv  get_availabl: prompt cache update took 0.02 ms
+0.26.380.071 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.380.099 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.27.073.027 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.27.138.087 I reasoning-budget: deactivated (natural end)
+0.27.269.343 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     728.42 ms /   424 tokens (    1.72 ms per token,   582.08 tokens per second)
+       eval time =     160.80 ms /     7 tokens (   22.97 ms per token,    43.53 tokens per second)
+      total time =     889.21 ms /   431 tokens
+0.27.269.422 I slot      release: id  0 | task 0 | stop processing: n_tokens = 430, truncated = 0
+0.27.269.432 I srv  update_slots: all slots are idle
+0.27.299.600 I srv  params_from_: Chat format: peg-native
+0.27.300.148 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.969 (> 0.100 thold), f_keep = 0.942
+0.27.300.612 I reasoning-budget: activated, budget=2147483647 tokens
+0.27.300.707 I slot launch_slot_: id  0 | task 9 | processing task, is_child = 0
+0.27.300.733 W slot update_slots: id  0 | task 9 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.27.300.737 I slot update_slots: id  0 | task 9 | Checking checkpoint with [419, 419] against 405...
+0.27.300.739 W slot update_slots: id  0 | task 9 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.27.300.746 W slot update_slots: id  0 | task 9 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.27.937.713 I slot create_check: id  0 | task 9 | created context checkpoint 1 of 32 (pos_min = 413, pos_max = 413, n_tokens = 414, size = 62.813 MiB)
+0.28.043.202 I reasoning-budget: deactivated (natural end)
+0.28.142.801 I slot print_timing: id  0 | task 9 | 
+prompt eval time =     701.16 ms /   418 tokens (    1.68 ms per token,   596.15 tokens per second)
+       eval time =     140.89 ms /     5 tokens (   28.18 ms per token,    35.49 tokens per second)
+      total time =     842.06 ms /   423 tokens
+0.28.142.910 I slot      release: id  0 | task 9 | stop processing: n_tokens = 422, truncated = 0
+0.28.142.959 I srv  update_slots: all slots are idle
+0.28.158.005 I srv  params_from_: Chat format: peg-native
+0.28.158.364 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.960
+0.28.158.578 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.158.608 I slot launch_slot_: id  0 | task 16 | processing task, is_child = 0
+0.28.158.616 W slot update_slots: id  0 | task 16 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.28.158.617 I slot update_slots: id  0 | task 16 | Checking checkpoint with [413, 413] against 405...
+0.28.158.618 W slot update_slots: id  0 | task 16 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.158.620 W slot update_slots: id  0 | task 16 | erased invalidated context checkpoint (pos_min = 413, pos_max = 413, n_tokens = 414, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.549.814 I slot create_check: id  0 | task 16 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.28.603.918 I reasoning-budget: deactivated (natural end)
+0.29.232.552 I slot print_timing: id  0 | task 16 | 
+prompt eval time =     422.03 ms /   421 tokens (    1.00 ms per token,   997.56 tokens per second)
+       eval time =     651.88 ms /    42 tokens (   15.52 ms per token,    64.43 tokens per second)
+      total time =    1073.91 ms /   463 tokens
+0.29.232.642 I slot      release: id  0 | task 16 | stop processing: n_tokens = 462, truncated = 0
+0.29.232.679 I srv  update_slots: all slots are idle
+0.29.259.355 I srv  params_from_: Chat format: peg-native
+0.29.259.776 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.951 (> 0.100 thold), f_keep = 0.879
+0.29.259.946 I reasoning-budget: activated, budget=2147483647 tokens
+0.29.259.949 I reasoning-budget: deactivated (natural end)
+0.29.259.987 I slot launch_slot_: id  0 | task 60 | processing task, is_child = 0
+0.29.259.997 W slot update_slots: id  0 | task 60 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.29.259.998 I slot update_slots: id  0 | task 60 | Checking checkpoint with [416, 416] against 406...
+0.29.260.000 W slot update_slots: id  0 | task 60 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.29.260.002 W slot update_slots: id  0 | task 60 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.29.608.781 I slot create_check: id  0 | task 60 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.29.705.934 I slot print_timing: id  0 | task 60 | 
+prompt eval time =     378.91 ms /   427 tokens (    0.89 ms per token,  1126.93 tokens per second)
+       eval time =      67.01 ms /     4 tokens (   16.75 ms per token,    59.69 tokens per second)
+      total time =     445.92 ms /   431 tokens
+0.29.706.025 I slot      release: id  0 | task 60 | stop processing: n_tokens = 430, truncated = 0
+0.29.706.063 I srv  update_slots: all slots are idle
+0.29.723.346 I srv  params_from_: Chat format: peg-native
+0.29.723.698 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.942
+0.29.723.878 I reasoning-budget: activated, budget=2147483647 tokens
+0.29.723.880 I reasoning-budget: deactivated (natural end)
+0.29.723.911 I slot launch_slot_: id  0 | task 66 | processing task, is_child = 0
+0.29.723.921 W slot update_slots: id  0 | task 66 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.29.723.922 I slot update_slots: id  0 | task 66 | Checking checkpoint with [422, 422] against 405...
+0.29.723.924 W slot update_slots: id  0 | task 66 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.29.723.926 W slot update_slots: id  0 | task 66 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.074.713 I slot create_check: id  0 | task 66 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.30.128.419 I slot print_timing: id  0 | task 66 | 
+prompt eval time =     380.89 ms /   421 tokens (    0.90 ms per token,  1105.32 tokens per second)
+       eval time =      23.59 ms /     2 tokens (   11.79 ms per token,    84.79 tokens per second)
+      total time =     404.48 ms /   423 tokens
+0.30.128.524 I slot      release: id  0 | task 66 | stop processing: n_tokens = 422, truncated = 0
+0.30.128.572 I srv  update_slots: all slots are idle
+0.30.150.200 I srv  params_from_: Chat format: peg-native
+0.30.150.557 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.960
+0.30.150.711 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.150.713 I reasoning-budget: deactivated (natural end)
+0.30.150.739 I slot launch_slot_: id  0 | task 70 | processing task, is_child = 0
+0.30.150.744 W slot update_slots: id  0 | task 70 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.30.150.745 I slot update_slots: id  0 | task 70 | Checking checkpoint with [416, 416] against 405...
+0.30.150.746 W slot update_slots: id  0 | task 70 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.150.752 W slot update_slots: id  0 | task 70 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.504.785 I slot create_check: id  0 | task 70 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.31.192.911 I slot print_timing: id  0 | task 70 | 
+prompt eval time =     383.87 ms /   424 tokens (    0.91 ms per token,  1104.53 tokens per second)
+       eval time =     658.28 ms /    39 tokens (   16.88 ms per token,    59.25 tokens per second)
+      total time =    1042.15 ms /   463 tokens
+0.31.192.995 I slot      release: id  0 | task 70 | stop processing: n_tokens = 462, truncated = 0
+0.31.193.032 I srv  update_slots: all slots are idle
+0.31.213.646 I srv  params_from_: Chat format: peg-native
+0.31.213.997 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.879
+0.31.214.182 I reasoning-budget: activated, budget=2147483647 tokens
+0.31.214.218 I slot launch_slot_: id  0 | task 111 | processing task, is_child = 0
+0.31.214.228 W slot update_slots: id  0 | task 111 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.31.214.230 I slot update_slots: id  0 | task 111 | Checking checkpoint with [419, 419] against 406...
+0.31.214.231 W slot update_slots: id  0 | task 111 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.31.214.234 W slot update_slots: id  0 | task 111 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.31.741.413 I slot create_check: id  0 | task 111 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.32.002.855 I reasoning-budget: deactivated (natural end)
+0.32.103.990 I slot print_timing: id  0 | task 111 | 
+prompt eval time =     558.28 ms /   425 tokens (    1.31 ms per token,   761.27 tokens per second)
+       eval time =     331.47 ms /    17 tokens (   19.50 ms per token,    51.29 tokens per second)
+      total time =     889.75 ms /   442 tokens
+0.32.104.074 I slot      release: id  0 | task 111 | stop processing: n_tokens = 441, truncated = 0
+0.32.104.109 I srv  update_slots: all slots are idle
+0.32.155.224 I srv  params_from_: Chat format: peg-native
+0.32.157.476 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.918
+0.32.157.999 I reasoning-budget: activated, budget=2147483647 tokens
+0.32.158.095 I slot launch_slot_: id  0 | task 130 | processing task, is_child = 0
+0.32.158.121 W slot update_slots: id  0 | task 130 | n_past = 405, slot.prompt.tokens.size() = 441, seq_id = 0, pos_min = 440, n_swa = 0
+0.32.158.123 I slot update_slots: id  0 | task 130 | Checking checkpoint with [420, 420] against 405...
+0.32.158.124 W slot update_slots: id  0 | task 130 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.32.158.130 W slot update_slots: id  0 | task 130 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.32.714.547 I slot create_check: id  0 | task 130 | created context checkpoint 1 of 32 (pos_min = 414, pos_max = 414, n_tokens = 415, size = 62.813 MiB)
+0.32.960.511 I reasoning-budget: deactivated (natural end)
+0.33.017.861 I slot print_timing: id  0 | task 130 | 
+prompt eval time =     601.42 ms /   419 tokens (    1.44 ms per token,   696.69 tokens per second)
+       eval time =     258.30 ms /    12 tokens (   21.53 ms per token,    46.46 tokens per second)
+      total time =     859.72 ms /   431 tokens
+0.33.018.052 I slot      release: id  0 | task 130 | stop processing: n_tokens = 430, truncated = 0
+0.33.018.108 I srv  update_slots: all slots are idle
+0.33.050.367 I srv  params_from_: Chat format: peg-native
+0.33.050.835 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.960 (> 0.100 thold), f_keep = 0.942
+0.33.051.413 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.051.498 I slot launch_slot_: id  0 | task 144 | processing task, is_child = 0
+0.33.051.517 W slot update_slots: id  0 | task 144 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.33.051.521 I slot update_slots: id  0 | task 144 | Checking checkpoint with [414, 414] against 405...
+0.33.051.522 W slot update_slots: id  0 | task 144 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.33.051.528 W slot update_slots: id  0 | task 144 | erased invalidated context checkpoint (pos_min = 414, pos_max = 414, n_tokens = 415, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.33.615.892 I slot create_check: id  0 | task 144 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.33.916.408 I reasoning-budget: deactivated (natural end)
+0.34.765.727 I slot print_timing: id  0 | task 144 | 
+prompt eval time =     597.52 ms /   422 tokens (    1.42 ms per token,   706.25 tokens per second)
+       eval time =    1116.68 ms /    53 tokens (   21.07 ms per token,    47.46 tokens per second)
+      total time =    1714.20 ms /   475 tokens
+0.34.765.808 I slot      release: id  0 | task 144 | stop processing: n_tokens = 474, truncated = 0
+0.34.765.835 I srv  update_slots: all slots are idle
+0.34.781.504 I srv  params_from_: Chat format: peg-native
+0.34.781.937 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.857
+0.34.782.155 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.782.201 I slot launch_slot_: id  0 | task 199 | processing task, is_child = 0
+0.34.782.213 W slot update_slots: id  0 | task 199 | n_past = 406, slot.prompt.tokens.size() = 474, seq_id = 0, pos_min = 473, n_swa = 0
+0.34.782.213 I slot update_slots: id  0 | task 199 | Checking checkpoint with [417, 417] against 406...
+0.34.782.214 W slot update_slots: id  0 | task 199 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.34.782.218 W slot update_slots: id  0 | task 199 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.312.464 I slot create_check: id  0 | task 199 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.35.382.494 I reasoning-budget: deactivated (natural end)
+0.35.500.672 I slot print_timing: id  0 | task 199 | 
+prompt eval time =     569.17 ms /   424 tokens (    1.34 ms per token,   744.95 tokens per second)
+       eval time =     149.28 ms /     7 tokens (   21.33 ms per token,    46.89 tokens per second)
+      total time =     718.45 ms /   431 tokens
+0.35.500.758 I slot      release: id  0 | task 199 | stop processing: n_tokens = 430, truncated = 0
+0.35.500.792 I srv  update_slots: all slots are idle
+0.35.523.836 I srv  params_from_: Chat format: peg-native
+0.35.524.275 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.969 (> 0.100 thold), f_keep = 0.942
+0.35.524.794 I reasoning-budget: activated, budget=2147483647 tokens
+0.35.524.873 I slot launch_slot_: id  0 | task 208 | processing task, is_child = 0
+0.35.524.891 W slot update_slots: id  0 | task 208 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.35.524.892 I slot update_slots: id  0 | task 208 | Checking checkpoint with [419, 419] against 405...
+0.35.524.894 W slot update_slots: id  0 | task 208 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.35.524.899 W slot update_slots: id  0 | task 208 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.142.960 I slot create_check: id  0 | task 208 | created context checkpoint 1 of 32 (pos_min = 413, pos_max = 413, n_tokens = 414, size = 62.813 MiB)
+0.36.260.514 I reasoning-budget: deactivated (natural end)
+0.36.391.695 I slot print_timing: id  0 | task 208 | 
+prompt eval time =     684.84 ms /   418 tokens (    1.64 ms per token,   610.36 tokens per second)
+       eval time =     181.96 ms /     5 tokens (   36.39 ms per token,    27.48 tokens per second)
+      total time =     866.79 ms /   423 tokens
+0.36.391.789 I slot      release: id  0 | task 208 | stop processing: n_tokens = 422, truncated = 0
+0.36.391.820 I srv  update_slots: all slots are idle
+0.36.404.024 I srv  params_from_: Chat format: peg-native
+0.36.404.587 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.960
+0.36.404.848 I reasoning-budget: activated, budget=2147483647 tokens
+0.36.404.897 I slot launch_slot_: id  0 | task 215 | processing task, is_child = 0
+0.36.404.910 W slot update_slots: id  0 | task 215 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.36.404.912 I slot update_slots: id  0 | task 215 | Checking checkpoint with [413, 413] against 405...
+0.36.404.913 W slot update_slots: id  0 | task 215 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.36.404.916 W slot update_slots: id  0 | task 215 | erased invalidated context checkpoint (pos_min = 413, pos_max = 413, n_tokens = 414, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.899.390 I slot create_check: id  0 | task 215 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.37.012.108 I reasoning-budget: deactivated (natural end)
+0.37.909.549 I slot print_timing: id  0 | task 215 | 
+prompt eval time =     555.94 ms /   421 tokens (    1.32 ms per token,   757.28 tokens per second)
+       eval time =     948.68 ms /    42 tokens (   22.59 ms per token,    44.27 tokens per second)
+      total time =    1504.62 ms /   463 tokens
+0.37.909.654 I slot      release: id  0 | task 215 | stop processing: n_tokens = 462, truncated = 0
+0.37.909.691 I srv  update_slots: all slots are idle
+0.37.937.856 I srv  params_from_: Chat format: peg-native
+0.37.938.240 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.951 (> 0.100 thold), f_keep = 0.879
+0.37.938.471 I reasoning-budget: activated, budget=2147483647 tokens
+0.37.938.517 I slot launch_slot_: id  0 | task 259 | processing task, is_child = 0
+0.37.938.529 W slot update_slots: id  0 | task 259 | n_past = 406, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.37.938.531 I slot update_slots: id  0 | task 259 | Checking checkpoint with [416, 416] against 406...
+0.37.938.532 W slot update_slots: id  0 | task 259 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.37.938.535 W slot update_slots: id  0 | task 259 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.38.478.969 I slot create_check: id  0 | task 259 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.38.599.473 I slot print_timing: id  0 | task 259 | 
+prompt eval time =     580.20 ms /   427 tokens (    1.36 ms per token,   735.95 tokens per second)
+       eval time =      80.65 ms /     4 tokens (   20.16 ms per token,    49.60 tokens per second)
+      total time =     660.85 ms /   431 tokens
+0.38.599.762 I slot      release: id  0 | task 259 | stop processing: n_tokens = 430, truncated = 0
+0.38.599.853 I srv  update_slots: all slots are idle
+0.38.629.562 I srv  params_from_: Chat format: peg-native
+0.38.630.012 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.942
+0.38.630.609 I reasoning-budget: activated, budget=2147483647 tokens
+0.38.630.693 I slot launch_slot_: id  0 | task 265 | processing task, is_child = 0
+0.38.630.711 W slot update_slots: id  0 | task 265 | n_past = 405, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.38.630.716 I slot update_slots: id  0 | task 265 | Checking checkpoint with [422, 422] against 405...
+0.38.630.719 W slot update_slots: id  0 | task 265 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.38.630.724 W slot update_slots: id  0 | task 265 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.39.188.051 I slot create_check: id  0 | task 265 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.39.260.024 I slot print_timing: id  0 | task 265 | 
+prompt eval time =     591.28 ms /   421 tokens (    1.40 ms per token,   712.01 tokens per second)
+       eval time =      37.99 ms /     2 tokens (   19.00 ms per token,    52.64 tokens per second)
+      total time =     629.28 ms /   423 tokens
+0.39.260.276 I slot      release: id  0 | task 265 | stop processing: n_tokens = 422, truncated = 0
+0.39.260.338 I srv  update_slots: all slots are idle
+0.39.312.544 I srv  params_from_: Chat format: peg-native
+0.39.314.833 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.960
+0.39.315.431 I reasoning-budget: activated, budget=2147483647 tokens
+0.39.315.516 I slot launch_slot_: id  0 | task 269 | processing task, is_child = 0
+0.39.315.539 W slot update_slots: id  0 | task 269 | n_past = 405, slot.prompt.tokens.size() = 422, seq_id = 0, pos_min = 421, n_swa = 0
+0.39.315.543 I slot update_slots: id  0 | task 269 | Checking checkpoint with [416, 416] against 405...
+0.39.315.545 W slot update_slots: id  0 | task 269 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.39.315.552 W slot update_slots: id  0 | task 269 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.39.893.482 I slot create_check: id  0 | task 269 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.40.793.357 I slot print_timing: id  0 | task 269 | 
+prompt eval time =     633.79 ms /   424 tokens (    1.49 ms per token,   668.99 tokens per second)
+       eval time =     844.02 ms /    39 tokens (   21.64 ms per token,    46.21 tokens per second)
+      total time =    1477.81 ms /   463 tokens
+0.40.793.457 I slot      release: id  0 | task 269 | stop processing: n_tokens = 462, truncated = 0
+0.40.793.487 I srv  update_slots: all slots are idle
+0.40.794.731 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106-tpl-medium.log

@@ -0,0 +1,294 @@
+0.00.105.454 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.105.464 I device_info:
+0.00.105.601 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 122408 MiB free)
+0.00.105.795 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.105.803 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.105.902 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.105.988 I srv          init: running without SSL
+0.00.106.040 I srv          init: using 31 threads for HTTP server
+0.00.106.042 I srv          init: the WebUI is disabled
+0.00.106.129 I srv         start: binding port with default address family
+0.00.107.354 I srv          main: loading model
+0.00.107.360 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.146.109 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.23.588.699 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.23.912.034 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.24.349.867 I srv    load_model: initializing slots, n_slots = 1
+0.24.567.364 W srv    load_model: speculative decoding will use checkpoints
+0.24.567.384 W common_speculative_init: no implementations specified for speculative decoding
+0.24.567.388 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.24.567.468 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.24.567.470 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.24.567.551 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.24.579.260 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.24.587.661 I srv          init: init: chat template, thinking = 1
+0.24.587.712 I srv          main: model loaded
+0.24.587.715 I srv          main: server is listening on http://127.0.0.1:18600
+0.24.587.720 I srv  update_slots: all slots are idle
+0.26.148.362 I srv  params_from_: Chat format: peg-native
+0.26.149.988 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.26.149.996 I srv  get_availabl: updating prompt cache
+0.26.150.007 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.26.150.015 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.26.150.018 I srv  get_availabl: prompt cache update took 0.02 ms
+0.26.150.786 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.150.816 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.26.871.165 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.27.133.131 I reasoning-budget: deactivated (natural end)
+0.28.001.263 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     773.86 ms /   421 tokens (    1.84 ms per token,   544.03 tokens per second)
+       eval time =    1076.54 ms /    48 tokens (   22.43 ms per token,    44.59 tokens per second)
+      total time =    1850.39 ms /   469 tokens
+0.28.001.472 I slot      release: id  0 | task 0 | stop processing: n_tokens = 468, truncated = 0
+0.28.001.492 I srv  update_slots: all slots are idle
+0.28.065.185 I srv  params_from_: Chat format: peg-native
+0.28.065.968 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.908 (> 0.100 thold), f_keep = 0.865
+0.28.066.222 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.066.277 I slot launch_slot_: id  0 | task 50 | processing task, is_child = 0
+0.28.066.290 W slot update_slots: id  0 | task 50 | n_past = 405, slot.prompt.tokens.size() = 468, seq_id = 0, pos_min = 467, n_swa = 0
+0.28.066.292 I slot update_slots: id  0 | task 50 | Checking checkpoint with [416, 416] against 405...
+0.28.066.293 W slot update_slots: id  0 | task 50 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.066.297 W slot update_slots: id  0 | task 50 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.663.976 I slot create_check: id  0 | task 50 | created context checkpoint 1 of 32 (pos_min = 441, pos_max = 441, n_tokens = 442, size = 62.813 MiB)
+0.29.408.376 I reasoning-budget: deactivated (natural end)
+0.30.568.909 I slot print_timing: id  0 | task 50 | 
+prompt eval time =     632.88 ms /   446 tokens (    1.42 ms per token,   704.71 tokens per second)
+       eval time =    1869.72 ms /    81 tokens (   23.08 ms per token,    43.32 tokens per second)
+      total time =    2502.60 ms /   527 tokens
+0.30.568.986 I slot      release: id  0 | task 50 | stop processing: n_tokens = 526, truncated = 0
+0.30.569.017 I srv  update_slots: all slots are idle
+0.30.584.452 I srv  params_from_: Chat format: peg-native
+0.30.584.982 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.770
+0.30.585.566 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.585.633 I slot launch_slot_: id  0 | task 133 | processing task, is_child = 0
+0.30.585.660 W slot update_slots: id  0 | task 133 | n_past = 405, slot.prompt.tokens.size() = 526, seq_id = 0, pos_min = 525, n_swa = 0
+0.30.585.663 I slot update_slots: id  0 | task 133 | Checking checkpoint with [441, 441] against 405...
+0.30.585.666 W slot update_slots: id  0 | task 133 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.585.671 W slot update_slots: id  0 | task 133 | erased invalidated context checkpoint (pos_min = 441, pos_max = 441, n_tokens = 442, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.31.126.980 I slot create_check: id  0 | task 133 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.31.358.672 I reasoning-budget: deactivated (natural end)
+0.32.155.583 I slot print_timing: id  0 | task 133 | 
+prompt eval time =     578.11 ms /   424 tokens (    1.36 ms per token,   733.42 tokens per second)
+       eval time =     991.80 ms /    48 tokens (   20.66 ms per token,    48.40 tokens per second)
+      total time =    1569.91 ms /   472 tokens
+0.32.155.659 I slot      release: id  0 | task 133 | stop processing: n_tokens = 471, truncated = 0
+0.32.155.694 I srv  update_slots: all slots are idle
+0.32.187.576 I srv  params_from_: Chat format: peg-native
+0.32.188.058 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.860
+0.32.188.654 I reasoning-budget: activated, budget=2147483647 tokens
+0.32.188.751 I slot launch_slot_: id  0 | task 183 | processing task, is_child = 0
+0.32.188.776 W slot update_slots: id  0 | task 183 | n_past = 405, slot.prompt.tokens.size() = 471, seq_id = 0, pos_min = 470, n_swa = 0
+0.32.188.780 I slot update_slots: id  0 | task 183 | Checking checkpoint with [419, 419] against 405...
+0.32.188.782 W slot update_slots: id  0 | task 183 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.32.188.790 W slot update_slots: id  0 | task 183 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.32.763.234 I slot create_check: id  0 | task 183 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.32.827.843 I reasoning-budget: deactivated (natural end)
+0.32.942.756 I slot print_timing: id  0 | task 183 | 
+prompt eval time =     610.58 ms /   424 tokens (    1.44 ms per token,   694.43 tokens per second)
+       eval time =     143.38 ms /     7 tokens (   20.48 ms per token,    48.82 tokens per second)
+      total time =     753.96 ms /   431 tokens
+0.32.942.940 I slot      release: id  0 | task 183 | stop processing: n_tokens = 430, truncated = 0
+0.32.942.992 I srv  update_slots: all slots are idle
+0.32.959.301 I srv  params_from_: Chat format: peg-native
+0.32.959.816 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.964 (> 0.100 thold), f_keep = 0.944
+0.32.960.404 I reasoning-budget: activated, budget=2147483647 tokens
+0.32.960.503 I slot launch_slot_: id  0 | task 192 | processing task, is_child = 0
+0.32.960.525 W slot update_slots: id  0 | task 192 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.32.960.529 I slot update_slots: id  0 | task 192 | Checking checkpoint with [419, 419] against 406...
+0.32.960.530 W slot update_slots: id  0 | task 192 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.32.960.538 W slot update_slots: id  0 | task 192 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.33.460.168 I slot create_check: id  0 | task 192 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.33.682.213 I reasoning-budget: deactivated (natural end)
+0.34.478.330 I slot print_timing: id  0 | task 192 | 
+prompt eval time =     535.13 ms /   421 tokens (    1.27 ms per token,   786.73 tokens per second)
+       eval time =     982.66 ms /    50 tokens (   19.65 ms per token,    50.88 tokens per second)
+      total time =    1517.79 ms /   471 tokens
+0.34.478.426 I slot      release: id  0 | task 192 | stop processing: n_tokens = 470, truncated = 0
+0.34.478.463 I srv  update_slots: all slots are idle
+0.34.496.222 I srv  params_from_: Chat format: peg-native
+0.34.496.734 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.942 (> 0.100 thold), f_keep = 1.000
+0.34.496.948 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.496.989 I slot launch_slot_: id  0 | task 244 | processing task, is_child = 0
+0.34.639.263 I slot create_check: id  0 | task 244 | created context checkpoint 2 of 32 (pos_min = 494, pos_max = 494, n_tokens = 495, size = 62.813 MiB)
+0.34.705.972 I reasoning-budget: deactivated (natural end)
+0.35.076.072 I slot print_timing: id  0 | task 244 | 
+prompt eval time =     177.70 ms /    29 tokens (    6.13 ms per token,   163.19 tokens per second)
+       eval time =     401.35 ms /    17 tokens (   23.61 ms per token,    42.36 tokens per second)
+      total time =     579.05 ms /    46 tokens
+0.35.076.179 I slot      release: id  0 | task 244 | stop processing: n_tokens = 515, truncated = 0
+0.35.076.214 I srv  update_slots: all slots are idle
+0.35.091.806 I srv  params_from_: Chat format: peg-native
+0.35.092.529 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.974 (> 0.100 thold), f_keep = 0.796
+0.35.093.006 I reasoning-budget: activated, budget=2147483647 tokens
+0.35.093.087 I slot launch_slot_: id  0 | task 263 | processing task, is_child = 0
+0.35.093.108 W slot update_slots: id  0 | task 263 | n_past = 410, slot.prompt.tokens.size() = 515, seq_id = 0, pos_min = 514, n_swa = 0
+0.35.093.110 I slot update_slots: id  0 | task 263 | Checking checkpoint with [494, 494] against 410...
+0.35.093.111 I slot update_slots: id  0 | task 263 | Checking checkpoint with [416, 416] against 410...
+0.35.093.113 W slot update_slots: id  0 | task 263 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.35.093.119 W slot update_slots: id  0 | task 263 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.094.217 W slot update_slots: id  0 | task 263 | erased invalidated context checkpoint (pos_min = 494, pos_max = 494, n_tokens = 495, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.35.626.770 I slot create_check: id  0 | task 263 | created context checkpoint 1 of 32 (pos_min = 416, pos_max = 416, n_tokens = 417, size = 62.813 MiB)
+0.36.142.008 I reasoning-budget: deactivated (natural end)
+0.36.985.537 I slot print_timing: id  0 | task 263 | 
+prompt eval time =     581.67 ms /   421 tokens (    1.38 ms per token,   723.78 tokens per second)
+       eval time =    1310.72 ms /    64 tokens (   20.48 ms per token,    48.83 tokens per second)
+      total time =    1892.39 ms /   485 tokens
+0.36.985.731 I slot      release: id  0 | task 263 | stop processing: n_tokens = 484, truncated = 0
+0.36.985.791 I srv  update_slots: all slots are idle
+0.37.040.565 I srv  params_from_: Chat format: peg-native
+0.37.041.116 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.938 (> 0.100 thold), f_keep = 0.837
+0.37.041.402 I reasoning-budget: activated, budget=2147483647 tokens
+0.37.041.512 I slot launch_slot_: id  0 | task 329 | processing task, is_child = 0
+0.37.041.525 W slot update_slots: id  0 | task 329 | n_past = 405, slot.prompt.tokens.size() = 484, seq_id = 0, pos_min = 483, n_swa = 0
+0.37.041.528 I slot update_slots: id  0 | task 329 | Checking checkpoint with [416, 416] against 405...
+0.37.041.528 W slot update_slots: id  0 | task 329 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.37.041.531 W slot update_slots: id  0 | task 329 | erased invalidated context checkpoint (pos_min = 416, pos_max = 416, n_tokens = 417, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.37.608.593 I slot create_check: id  0 | task 329 | created context checkpoint 1 of 32 (pos_min = 427, pos_max = 427, n_tokens = 428, size = 62.813 MiB)
+0.38.415.584 I reasoning-budget: deactivated (natural end)
+0.40.014.106 I slot print_timing: id  0 | task 329 | n_decoded =    100, tg =  42.63 t/s
+0.40.267.618 I slot print_timing: id  0 | task 329 | 
+prompt eval time =     626.94 ms /   432 tokens (    1.45 ms per token,   689.06 tokens per second)
+       eval time =    2599.14 ms /   113 tokens (   23.00 ms per token,    43.48 tokens per second)
+      total time =    3226.08 ms /   545 tokens
+0.40.267.691 I slot      release: id  0 | task 329 | stop processing: n_tokens = 544, truncated = 0
+0.40.267.739 I srv  update_slots: all slots are idle
+0.40.291.843 I srv  params_from_: Chat format: peg-native
+0.40.292.295 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.744
+0.40.292.805 I reasoning-budget: activated, budget=2147483647 tokens
+0.40.292.809 I reasoning-budget: deactivated (natural end)
+0.40.292.882 I slot launch_slot_: id  0 | task 444 | processing task, is_child = 0
+0.40.292.903 W slot update_slots: id  0 | task 444 | n_past = 405, slot.prompt.tokens.size() = 544, seq_id = 0, pos_min = 543, n_swa = 0
+0.40.292.906 I slot update_slots: id  0 | task 444 | Checking checkpoint with [427, 427] against 405...
+0.40.292.907 W slot update_slots: id  0 | task 444 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.40.292.912 W slot update_slots: id  0 | task 444 | erased invalidated context checkpoint (pos_min = 427, pos_max = 427, n_tokens = 428, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.40.808.790 I slot create_check: id  0 | task 444 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.41.645.500 I slot print_timing: id  0 | task 444 | 
+prompt eval time =     551.51 ms /   424 tokens (    1.30 ms per token,   768.79 tokens per second)
+       eval time =     801.07 ms /    39 tokens (   20.54 ms per token,    48.68 tokens per second)
+      total time =    1352.58 ms /   463 tokens
+0.41.645.589 I slot      release: id  0 | task 444 | stop processing: n_tokens = 462, truncated = 0
+0.41.645.620 I srv  update_slots: all slots are idle
+0.41.661.261 I srv  params_from_: Chat format: peg-native
+0.41.661.789 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.902 (> 0.100 thold), f_keep = 0.877
+0.41.662.003 I reasoning-budget: activated, budget=2147483647 tokens
+0.41.662.005 I reasoning-budget: deactivated (natural end)
+0.41.662.045 I slot launch_slot_: id  0 | task 485 | processing task, is_child = 0
+0.41.662.055 W slot update_slots: id  0 | task 485 | n_past = 405, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.41.662.056 I slot update_slots: id  0 | task 485 | Checking checkpoint with [419, 419] against 405...
+0.41.662.057 W slot update_slots: id  0 | task 485 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.41.662.060 W slot update_slots: id  0 | task 485 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.42.169.075 I slot create_check: id  0 | task 485 | created context checkpoint 1 of 32 (pos_min = 444, pos_max = 444, n_tokens = 445, size = 62.813 MiB)
+0.44.038.103 I slot print_timing: id  0 | task 485 | 
+prompt eval time =     548.61 ms /   449 tokens (    1.22 ms per token,   818.43 tokens per second)
+       eval time =    1827.42 ms /    87 tokens (   21.00 ms per token,    47.61 tokens per second)
+      total time =    2376.03 ms /   536 tokens
+0.44.038.196 I slot      release: id  0 | task 485 | stop processing: n_tokens = 535, truncated = 0
+0.44.038.227 I srv  update_slots: all slots are idle
+0.44.070.792 I srv  params_from_: Chat format: peg-native
+0.44.071.263 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.757
+0.44.071.950 I reasoning-budget: activated, budget=2147483647 tokens
+0.44.071.955 I reasoning-budget: deactivated (natural end)
+0.44.072.053 I slot launch_slot_: id  0 | task 574 | processing task, is_child = 0
+0.44.072.077 W slot update_slots: id  0 | task 574 | n_past = 405, slot.prompt.tokens.size() = 535, seq_id = 0, pos_min = 534, n_swa = 0
+0.44.072.081 I slot update_slots: id  0 | task 574 | Checking checkpoint with [444, 444] against 405...
+0.44.072.083 W slot update_slots: id  0 | task 574 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.44.072.088 W slot update_slots: id  0 | task 574 | erased invalidated context checkpoint (pos_min = 444, pos_max = 444, n_tokens = 445, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.44.616.066 I slot create_check: id  0 | task 574 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.45.562.940 I slot print_timing: id  0 | task 574 | 
+prompt eval time =     603.10 ms /   427 tokens (    1.41 ms per token,   708.01 tokens per second)
+       eval time =     887.75 ms /    39 tokens (   22.76 ms per token,    43.93 tokens per second)
+      total time =    1490.84 ms /   466 tokens
+0.45.563.042 I slot      release: id  0 | task 574 | stop processing: n_tokens = 465, truncated = 0
+0.45.563.082 I srv  update_slots: all slots are idle
+0.45.579.534 I srv  params_from_: Chat format: peg-native
+0.45.580.094 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.871
+0.45.580.642 I reasoning-budget: activated, budget=2147483647 tokens
+0.45.580.646 I reasoning-budget: deactivated (natural end)
+0.45.582.811 I slot launch_slot_: id  0 | task 615 | processing task, is_child = 0
+0.45.582.848 W slot update_slots: id  0 | task 615 | n_past = 405, slot.prompt.tokens.size() = 465, seq_id = 0, pos_min = 464, n_swa = 0
+0.45.582.849 I slot update_slots: id  0 | task 615 | Checking checkpoint with [422, 422] against 405...
+0.45.582.851 W slot update_slots: id  0 | task 615 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.45.582.860 W slot update_slots: id  0 | task 615 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.46.041.817 I slot create_check: id  0 | task 615 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.46.171.798 I slot print_timing: id  0 | task 615 | 
+prompt eval time =     507.02 ms /   427 tokens (    1.19 ms per token,   842.17 tokens per second)
+       eval time =      81.92 ms /     4 tokens (   20.48 ms per token,    48.83 tokens per second)
+      total time =     588.94 ms /   431 tokens
+0.46.171.890 I slot      release: id  0 | task 615 | stop processing: n_tokens = 430, truncated = 0
+0.46.171.920 I srv  update_slots: all slots are idle
+0.46.188.295 I srv  params_from_: Chat format: peg-native
+0.46.188.729 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.944
+0.46.189.261 I reasoning-budget: activated, budget=2147483647 tokens
+0.46.189.264 I reasoning-budget: deactivated (natural end)
+0.46.189.348 I slot launch_slot_: id  0 | task 621 | processing task, is_child = 0
+0.46.189.367 W slot update_slots: id  0 | task 621 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.46.189.370 I slot update_slots: id  0 | task 621 | Checking checkpoint with [422, 422] against 406...
+0.46.189.372 W slot update_slots: id  0 | task 621 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.46.189.377 W slot update_slots: id  0 | task 621 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.46.716.804 I slot create_check: id  0 | task 621 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.47.566.188 I slot print_timing: id  0 | task 621 | 
+prompt eval time =     563.24 ms /   424 tokens (    1.33 ms per token,   752.78 tokens per second)
+       eval time =     813.55 ms /    40 tokens (   20.34 ms per token,    49.17 tokens per second)
+      total time =    1376.79 ms /   464 tokens
+0.47.566.402 I slot      release: id  0 | task 621 | stop processing: n_tokens = 463, truncated = 0
+0.47.566.467 I srv  update_slots: all slots are idle
+0.47.589.246 I srv  params_from_: Chat format: peg-native
+0.47.589.845 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 1.000
+0.47.590.510 I reasoning-budget: activated, budget=2147483647 tokens
+0.47.590.517 I reasoning-budget: deactivated (natural end)
+0.47.590.637 I slot launch_slot_: id  0 | task 663 | processing task, is_child = 0
+0.47.740.622 I slot create_check: id  0 | task 663 | created context checkpoint 2 of 32 (pos_min = 490, pos_max = 490, n_tokens = 491, size = 62.813 MiB)
+0.48.115.507 I slot print_timing: id  0 | task 663 | 
+prompt eval time =     197.73 ms /    32 tokens (    6.18 ms per token,   161.84 tokens per second)
+       eval time =     327.11 ms /    15 tokens (   21.81 ms per token,    45.86 tokens per second)
+      total time =     524.83 ms /    47 tokens
+0.48.115.614 I slot      release: id  0 | task 663 | stop processing: n_tokens = 509, truncated = 0
+0.48.115.648 I srv  update_slots: all slots are idle
+0.48.182.251 I srv  params_from_: Chat format: peg-native
+0.48.182.770 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.806
+0.48.183.002 I reasoning-budget: activated, budget=2147483647 tokens
+0.48.183.003 I reasoning-budget: deactivated (natural end)
+0.48.183.055 I slot launch_slot_: id  0 | task 680 | processing task, is_child = 0
+0.48.183.066 W slot update_slots: id  0 | task 680 | n_past = 410, slot.prompt.tokens.size() = 509, seq_id = 0, pos_min = 508, n_swa = 0
+0.48.183.067 I slot update_slots: id  0 | task 680 | Checking checkpoint with [490, 490] against 410...
+0.48.183.067 I slot update_slots: id  0 | task 680 | Checking checkpoint with [419, 419] against 410...
+0.48.183.069 W slot update_slots: id  0 | task 680 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.48.183.072 W slot update_slots: id  0 | task 680 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.48.184.972 W slot update_slots: id  0 | task 680 | erased invalidated context checkpoint (pos_min = 490, pos_max = 490, n_tokens = 491, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.48.742.624 I slot create_check: id  0 | task 680 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.49.641.904 I slot print_timing: id  0 | task 680 | 
+prompt eval time =     599.46 ms /   424 tokens (    1.41 ms per token,   707.31 tokens per second)
+       eval time =     859.36 ms /    40 tokens (   21.48 ms per token,    46.55 tokens per second)
+      total time =    1458.81 ms /   464 tokens
+0.49.641.977 I slot      release: id  0 | task 680 | stop processing: n_tokens = 463, truncated = 0
+0.49.642.005 I srv  update_slots: all slots are idle
+0.49.654.625 I srv  params_from_: Chat format: peg-native
+0.49.655.158 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.931 (> 0.100 thold), f_keep = 0.875
+0.49.655.408 I reasoning-budget: activated, budget=2147483647 tokens
+0.49.655.411 I reasoning-budget: deactivated (natural end)
+0.49.655.459 I slot launch_slot_: id  0 | task 722 | processing task, is_child = 0
+0.49.655.469 W slot update_slots: id  0 | task 722 | n_past = 405, slot.prompt.tokens.size() = 463, seq_id = 0, pos_min = 462, n_swa = 0
+0.49.655.470 I slot update_slots: id  0 | task 722 | Checking checkpoint with [419, 419] against 405...
+0.49.655.470 W slot update_slots: id  0 | task 722 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.49.655.472 W slot update_slots: id  0 | task 722 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.50.147.360 I slot create_check: id  0 | task 722 | created context checkpoint 1 of 32 (pos_min = 430, pos_max = 430, n_tokens = 431, size = 62.813 MiB)
+0.51.806.641 I slot print_timing: id  0 | task 722 | 
+prompt eval time =     536.58 ms /   435 tokens (    1.23 ms per token,   810.69 tokens per second)
+       eval time =    1614.55 ms /    80 tokens (   20.18 ms per token,    49.55 tokens per second)
+      total time =    2151.13 ms /   515 tokens
+0.51.807.033 I slot      release: id  0 | task 722 | stop processing: n_tokens = 514, truncated = 0
+0.51.807.128 I srv  update_slots: all slots are idle
+0.51.808.793 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-tools-q106.log

@@ -0,0 +1,302 @@
+0.00.039.657 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.039.660 I device_info:
+0.00.039.707 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 125232 MiB free)
+0.00.039.793 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131926 MiB free)
+0.00.039.797 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.039.840 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.039.855 I srv          init: running without SSL
+0.00.039.876 I srv          init: using 31 threads for HTTP server
+0.00.039.877 I srv          init: the WebUI is disabled
+0.00.039.934 I srv         start: binding port with default address family
+0.00.041.085 I srv          main: loading model
+0.00.041.087 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.075.504 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.409.072 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.574.487 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.774.364 I srv    load_model: initializing slots, n_slots = 1
+0.20.940.846 W srv    load_model: speculative decoding will use checkpoints
+0.20.940.851 W common_speculative_init: no implementations specified for speculative decoding
+0.20.940.852 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.20.940.886 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.20.940.894 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.20.940.907 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.20.950.043 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.20.957.344 I srv          init: init: chat template, thinking = 1
+0.20.957.368 I srv          main: model loaded
+0.20.957.372 I srv          main: server is listening on http://127.0.0.1:18600
+0.20.957.374 I srv  update_slots: all slots are idle
+0.22.001.528 I srv  params_from_: Chat format: peg-native
+0.22.001.842 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.22.001.845 I srv  get_availabl: updating prompt cache
+0.22.001.851 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.22.001.856 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.22.001.858 I srv  get_availabl: prompt cache update took 0.01 ms
+0.22.010.813 I reasoning-budget: activated, budget=2147483647 tokens
+0.22.010.830 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.22.354.924 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.22.608.502 I reasoning-budget: deactivated (natural end)
+0.23.202.843 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     372.12 ms /   422 tokens (    0.88 ms per token,  1134.04 tokens per second)
+       eval time =     819.87 ms /    55 tokens (   14.91 ms per token,    67.08 tokens per second)
+      total time =    1191.99 ms /   477 tokens
+0.23.202.893 I slot      release: id  0 | task 0 | stop processing: n_tokens = 476, truncated = 0
+0.23.202.897 I srv  update_slots: all slots are idle
+0.23.213.207 I srv  params_from_: Chat format: peg-native
+0.23.213.731 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.906 (> 0.100 thold), f_keep = 0.851
+0.23.213.930 I reasoning-budget: activated, budget=2147483647 tokens
+0.23.213.965 I slot launch_slot_: id  0 | task 57 | processing task, is_child = 0
+0.23.213.971 W slot update_slots: id  0 | task 57 | n_past = 405, slot.prompt.tokens.size() = 476, seq_id = 0, pos_min = 475, n_swa = 0
+0.23.213.971 I slot update_slots: id  0 | task 57 | Checking checkpoint with [417, 417] against 405...
+0.23.213.973 W slot update_slots: id  0 | task 57 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.23.213.975 W slot update_slots: id  0 | task 57 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.23.551.876 I slot create_check: id  0 | task 57 | created context checkpoint 1 of 32 (pos_min = 442, pos_max = 442, n_tokens = 443, size = 62.813 MiB)
+0.24.213.049 I reasoning-budget: deactivated (natural end)
+0.24.949.926 I slot print_timing: id  0 | task 57 | 
+prompt eval time =     365.53 ms /   447 tokens (    0.82 ms per token,  1222.88 tokens per second)
+       eval time =    1370.41 ms /    92 tokens (   14.90 ms per token,    67.13 tokens per second)
+      total time =    1735.95 ms /   539 tokens
+0.24.949.971 I slot      release: id  0 | task 57 | stop processing: n_tokens = 538, truncated = 0
+0.24.949.992 I srv  update_slots: all slots are idle
+0.24.959.419 I srv  params_from_: Chat format: peg-native
+0.24.959.740 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.753
+0.24.959.952 I reasoning-budget: activated, budget=2147483647 tokens
+0.24.959.981 I slot launch_slot_: id  0 | task 151 | processing task, is_child = 0
+0.24.959.987 W slot update_slots: id  0 | task 151 | n_past = 405, slot.prompt.tokens.size() = 538, seq_id = 0, pos_min = 537, n_swa = 0
+0.24.959.987 I slot update_slots: id  0 | task 151 | Checking checkpoint with [442, 442] against 405...
+0.24.959.988 W slot update_slots: id  0 | task 151 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.24.959.991 W slot update_slots: id  0 | task 151 | erased invalidated context checkpoint (pos_min = 442, pos_max = 442, n_tokens = 443, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.25.284.086 I slot create_check: id  0 | task 151 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.25.471.508 I reasoning-budget: deactivated (natural end)
+0.26.064.107 I slot print_timing: id  0 | task 151 | 
+prompt eval time =     351.76 ms /   425 tokens (    0.83 ms per token,  1208.20 tokens per second)
+       eval time =     752.35 ms /    51 tokens (   14.75 ms per token,    67.79 tokens per second)
+      total time =    1104.11 ms /   476 tokens
+0.26.064.151 I slot      release: id  0 | task 151 | stop processing: n_tokens = 475, truncated = 0
+0.26.064.171 I srv  update_slots: all slots are idle
+0.26.074.471 I srv  params_from_: Chat format: peg-native
+0.26.074.789 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.953 (> 0.100 thold), f_keep = 0.853
+0.26.074.999 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.075.029 I slot launch_slot_: id  0 | task 204 | processing task, is_child = 0
+0.26.075.034 W slot update_slots: id  0 | task 204 | n_past = 405, slot.prompt.tokens.size() = 475, seq_id = 0, pos_min = 474, n_swa = 0
+0.26.075.045 I slot update_slots: id  0 | task 204 | Checking checkpoint with [420, 420] against 405...
+0.26.075.048 W slot update_slots: id  0 | task 204 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.26.075.050 W slot update_slots: id  0 | task 204 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.26.399.124 I slot create_check: id  0 | task 204 | created context checkpoint 1 of 32 (pos_min = 420, pos_max = 420, n_tokens = 421, size = 62.813 MiB)
+0.26.601.525 I reasoning-budget: deactivated (natural end)
+0.26.675.559 I slot print_timing: id  0 | task 204 | 
+prompt eval time =     351.60 ms /   425 tokens (    0.83 ms per token,  1208.76 tokens per second)
+       eval time =     248.92 ms /    17 tokens (   14.64 ms per token,    68.30 tokens per second)
+      total time =     600.52 ms /   442 tokens
+0.26.675.604 I slot      release: id  0 | task 204 | stop processing: n_tokens = 441, truncated = 0
+0.26.675.625 I srv  update_slots: all slots are idle
+0.26.685.901 I srv  params_from_: Chat format: peg-native
+0.26.686.221 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.962 (> 0.100 thold), f_keep = 0.921
+0.26.686.435 I reasoning-budget: activated, budget=2147483647 tokens
+0.26.686.468 I slot launch_slot_: id  0 | task 223 | processing task, is_child = 0
+0.26.686.473 W slot update_slots: id  0 | task 223 | n_past = 406, slot.prompt.tokens.size() = 441, seq_id = 0, pos_min = 440, n_swa = 0
+0.26.686.474 I slot update_slots: id  0 | task 223 | Checking checkpoint with [420, 420] against 406...
+0.26.686.475 W slot update_slots: id  0 | task 223 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.26.686.476 W slot update_slots: id  0 | task 223 | erased invalidated context checkpoint (pos_min = 420, pos_max = 420, n_tokens = 421, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.27.011.819 I slot create_check: id  0 | task 223 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.27.481.528 I reasoning-budget: deactivated (natural end)
+0.28.088.823 I slot print_timing: id  0 | task 223 | 
+prompt eval time =     353.00 ms /   422 tokens (    0.84 ms per token,  1195.46 tokens per second)
+       eval time =    1049.34 ms /    71 tokens (   14.78 ms per token,    67.66 tokens per second)
+      total time =    1402.34 ms /   493 tokens
+0.28.088.866 I slot      release: id  0 | task 223 | stop processing: n_tokens = 492, truncated = 0
+0.28.088.892 I srv  update_slots: all slots are idle
+0.28.099.105 I srv  params_from_: Chat format: peg-native
+0.28.099.413 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.943 (> 0.100 thold), f_keep = 1.000
+0.28.099.624 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.099.652 I slot launch_slot_: id  0 | task 296 | processing task, is_child = 0
+0.28.109.016 I slot create_check: id  0 | task 296 | created context checkpoint 2 of 32 (pos_min = 491, pos_max = 491, n_tokens = 492, size = 62.813 MiB)
+0.28.638.379 I reasoning-budget: deactivated (natural end)
+0.28.905.201 I slot print_timing: id  0 | task 296 | 
+prompt eval time =     111.40 ms /    30 tokens (    3.71 ms per token,   269.30 tokens per second)
+       eval time =     694.13 ms /    47 tokens (   14.77 ms per token,    67.71 tokens per second)
+      total time =     805.53 ms /    77 tokens
+0.28.905.251 I slot      release: id  0 | task 296 | stop processing: n_tokens = 568, truncated = 0
+0.28.905.277 I srv  update_slots: all slots are idle
+0.28.915.951 I srv  params_from_: Chat format: peg-native
+0.28.916.241 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.972 (> 0.100 thold), f_keep = 0.722
+0.28.916.410 I reasoning-budget: activated, budget=2147483647 tokens
+0.28.916.437 I slot launch_slot_: id  0 | task 345 | processing task, is_child = 0
+0.28.916.443 W slot update_slots: id  0 | task 345 | n_past = 410, slot.prompt.tokens.size() = 568, seq_id = 0, pos_min = 567, n_swa = 0
+0.28.916.454 I slot update_slots: id  0 | task 345 | Checking checkpoint with [491, 491] against 410...
+0.28.916.456 I slot update_slots: id  0 | task 345 | Checking checkpoint with [417, 417] against 410...
+0.28.916.457 W slot update_slots: id  0 | task 345 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.28.916.459 W slot update_slots: id  0 | task 345 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.28.918.102 W slot update_slots: id  0 | task 345 | erased invalidated context checkpoint (pos_min = 491, pos_max = 491, n_tokens = 492, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.29.241.970 I slot create_check: id  0 | task 345 | created context checkpoint 1 of 32 (pos_min = 417, pos_max = 417, n_tokens = 418, size = 62.813 MiB)
+0.29.578.948 I reasoning-budget: deactivated (natural end)
+0.30.187.657 I slot print_timing: id  0 | task 345 | 
+prompt eval time =     354.33 ms /   422 tokens (    0.84 ms per token,  1190.98 tokens per second)
+       eval time =     916.87 ms /    62 tokens (   14.79 ms per token,    67.62 tokens per second)
+      total time =    1271.20 ms /   484 tokens
+0.30.187.698 I slot      release: id  0 | task 345 | stop processing: n_tokens = 483, truncated = 0
+0.30.187.716 I srv  update_slots: all slots are idle
+0.30.197.642 I srv  params_from_: Chat format: peg-native
+0.30.197.942 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 0.839
+0.30.198.086 I reasoning-budget: activated, budget=2147483647 tokens
+0.30.198.110 I slot launch_slot_: id  0 | task 409 | processing task, is_child = 0
+0.30.198.115 W slot update_slots: id  0 | task 409 | n_past = 405, slot.prompt.tokens.size() = 483, seq_id = 0, pos_min = 482, n_swa = 0
+0.30.198.116 I slot update_slots: id  0 | task 409 | Checking checkpoint with [417, 417] against 405...
+0.30.198.117 W slot update_slots: id  0 | task 409 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.30.198.120 W slot update_slots: id  0 | task 409 | erased invalidated context checkpoint (pos_min = 417, pos_max = 417, n_tokens = 418, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.30.540.883 I slot create_check: id  0 | task 409 | created context checkpoint 1 of 32 (pos_min = 428, pos_max = 428, n_tokens = 429, size = 62.813 MiB)
+0.32.074.181 I reasoning-budget: deactivated (natural end)
+0.32.119.135 I slot print_timing: id  0 | task 409 | n_decoded =    100, tg =  64.60 t/s
+0.33.280.892 I slot print_timing: id  0 | task 409 | 
+prompt eval time =     373.02 ms /   433 tokens (    0.86 ms per token,  1160.78 tokens per second)
+       eval time =    2709.74 ms /   178 tokens (   15.22 ms per token,    65.69 tokens per second)
+      total time =    3082.77 ms /   611 tokens
+0.33.280.942 I slot      release: id  0 | task 409 | stop processing: n_tokens = 610, truncated = 0
+0.33.280.983 I srv  update_slots: all slots are idle
+0.33.291.660 I srv  params_from_: Chat format: peg-native
+0.33.291.953 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.955 (> 0.100 thold), f_keep = 0.664
+0.33.292.103 I reasoning-budget: activated, budget=2147483647 tokens
+0.33.292.127 I slot launch_slot_: id  0 | task 589 | processing task, is_child = 0
+0.33.292.133 W slot update_slots: id  0 | task 589 | n_past = 405, slot.prompt.tokens.size() = 610, seq_id = 0, pos_min = 609, n_swa = 0
+0.33.292.133 I slot update_slots: id  0 | task 589 | Checking checkpoint with [428, 428] against 405...
+0.33.292.134 W slot update_slots: id  0 | task 589 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.33.292.136 W slot update_slots: id  0 | task 589 | erased invalidated context checkpoint (pos_min = 428, pos_max = 428, n_tokens = 429, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.33.632.176 I slot create_check: id  0 | task 589 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.34.264.101 I slot print_timing: id  0 | task 589 | 
+prompt eval time =     370.43 ms /   424 tokens (    0.87 ms per token,  1144.62 tokens per second)
+       eval time =     601.52 ms /    39 tokens (   15.42 ms per token,    64.84 tokens per second)
+      total time =     971.94 ms /   463 tokens
+0.34.264.186 I slot      release: id  0 | task 589 | stop processing: n_tokens = 462, truncated = 0
+0.34.264.222 I srv  update_slots: all slots are idle
+0.34.291.838 I srv  params_from_: Chat format: peg-native
+0.34.292.413 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.902 (> 0.100 thold), f_keep = 0.877
+0.34.292.609 I reasoning-budget: activated, budget=2147483647 tokens
+0.34.292.640 I slot launch_slot_: id  0 | task 630 | processing task, is_child = 0
+0.34.292.647 W slot update_slots: id  0 | task 630 | n_past = 405, slot.prompt.tokens.size() = 462, seq_id = 0, pos_min = 461, n_swa = 0
+0.34.292.648 I slot update_slots: id  0 | task 630 | Checking checkpoint with [419, 419] against 405...
+0.34.292.649 W slot update_slots: id  0 | task 630 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.34.292.653 W slot update_slots: id  0 | task 630 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.34.652.933 I slot create_check: id  0 | task 630 | created context checkpoint 1 of 32 (pos_min = 444, pos_max = 444, n_tokens = 445, size = 62.813 MiB)
+0.36.023.509 I slot print_timing: id  0 | task 630 | 
+prompt eval time =     390.02 ms /   449 tokens (    0.87 ms per token,  1151.23 tokens per second)
+       eval time =    1340.82 ms /    86 tokens (   15.59 ms per token,    64.14 tokens per second)
+      total time =    1730.84 ms /   535 tokens
+0.36.023.602 I slot      release: id  0 | task 630 | stop processing: n_tokens = 534, truncated = 0
+0.36.023.637 I srv  update_slots: all slots are idle
+0.36.023.710 W common_chat_peg_parse: unparsed peg-native output: <tool_call>
+<function=create_event>
+<parameter=attendees>
+["ana@x.io", "bo@x.io"]
+</parameter>
+<parameter=when>
+{"date": "2026-10-02", "time": "14:00"}
+</parameter>
+<parameter=title>
+Design review
+</parameter>
+</function>
+</tool_call>
+0.36.025.846 W srv          stop: cancel task, id_task = 630
+0.36.025.894 I srv  update_slots: all slots are idle
+0.36.025.957 W srv    operator(): got exception: {"error":{"code":500,"message":"The model produced output that does not match the expected peg-native format","type":"server_error"}}
+0.36.045.619 I srv  params_from_: Chat format: peg-native
+0.36.045.957 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.758
+0.36.046.107 I reasoning-budget: activated, budget=2147483647 tokens
+0.36.046.133 I slot launch_slot_: id  0 | task 719 | processing task, is_child = 0
+0.36.046.138 W slot update_slots: id  0 | task 719 | n_past = 405, slot.prompt.tokens.size() = 534, seq_id = 0, pos_min = 533, n_swa = 0
+0.36.046.139 I slot update_slots: id  0 | task 719 | Checking checkpoint with [444, 444] against 405...
+0.36.046.142 W slot update_slots: id  0 | task 719 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.36.046.144 W slot update_slots: id  0 | task 719 | erased invalidated context checkpoint (pos_min = 444, pos_max = 444, n_tokens = 445, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.36.387.995 I slot create_check: id  0 | task 719 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.37.019.525 I slot print_timing: id  0 | task 719 | 
+prompt eval time =     371.74 ms /   427 tokens (    0.87 ms per token,  1148.64 tokens per second)
+       eval time =     601.62 ms /    39 tokens (   15.43 ms per token,    64.82 tokens per second)
+      total time =     973.36 ms /   466 tokens
+0.37.019.612 I slot      release: id  0 | task 719 | stop processing: n_tokens = 465, truncated = 0
+0.37.019.648 I srv  update_slots: all slots are idle
+0.37.047.661 I srv  params_from_: Chat format: peg-native
+0.37.048.192 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.948 (> 0.100 thold), f_keep = 0.871
+0.37.048.354 I reasoning-budget: activated, budget=2147483647 tokens
+0.37.048.382 I slot launch_slot_: id  0 | task 760 | processing task, is_child = 0
+0.37.048.388 W slot update_slots: id  0 | task 760 | n_past = 405, slot.prompt.tokens.size() = 465, seq_id = 0, pos_min = 464, n_swa = 0
+0.37.048.390 I slot update_slots: id  0 | task 760 | Checking checkpoint with [422, 422] against 405...
+0.37.048.391 W slot update_slots: id  0 | task 760 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.37.048.393 W slot update_slots: id  0 | task 760 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.37.396.973 I slot create_check: id  0 | task 760 | created context checkpoint 1 of 32 (pos_min = 422, pos_max = 422, n_tokens = 423, size = 62.813 MiB)
+0.37.495.139 I slot print_timing: id  0 | task 760 | 
+prompt eval time =     378.68 ms /   427 tokens (    0.89 ms per token,  1127.60 tokens per second)
+       eval time =      68.05 ms /     4 tokens (   17.01 ms per token,    58.78 tokens per second)
+      total time =     446.73 ms /   431 tokens
+0.37.495.225 I slot      release: id  0 | task 760 | stop processing: n_tokens = 430, truncated = 0
+0.37.495.261 I srv  update_slots: all slots are idle
+0.37.511.484 I srv  params_from_: Chat format: peg-native
+0.37.511.815 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.958 (> 0.100 thold), f_keep = 0.944
+0.37.511.971 I reasoning-budget: activated, budget=2147483647 tokens
+0.37.512.000 I slot launch_slot_: id  0 | task 766 | processing task, is_child = 0
+0.37.512.007 W slot update_slots: id  0 | task 766 | n_past = 406, slot.prompt.tokens.size() = 430, seq_id = 0, pos_min = 429, n_swa = 0
+0.37.512.008 I slot update_slots: id  0 | task 766 | Checking checkpoint with [422, 422] against 406...
+0.37.512.009 W slot update_slots: id  0 | task 766 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.37.512.012 W slot update_slots: id  0 | task 766 | erased invalidated context checkpoint (pos_min = 422, pos_max = 422, n_tokens = 423, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.37.860.025 I slot create_check: id  0 | task 766 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.38.509.296 I slot print_timing: id  0 | task 766 | 
+prompt eval time =     377.88 ms /   424 tokens (    0.89 ms per token,  1122.06 tokens per second)
+       eval time =     619.39 ms /    40 tokens (   15.48 ms per token,    64.58 tokens per second)
+      total time =     997.27 ms /   464 tokens
+0.38.509.383 I slot      release: id  0 | task 766 | stop processing: n_tokens = 463, truncated = 0
+0.38.509.426 I srv  update_slots: all slots are idle
+0.38.537.775 I srv  params_from_: Chat format: peg-native
+0.38.538.231 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.935 (> 0.100 thold), f_keep = 1.000
+0.38.538.382 I reasoning-budget: activated, budget=2147483647 tokens
+0.38.538.411 I slot launch_slot_: id  0 | task 808 | processing task, is_child = 0
+0.38.628.133 I slot create_check: id  0 | task 808 | created context checkpoint 2 of 32 (pos_min = 490, pos_max = 490, n_tokens = 491, size = 62.813 MiB)
+0.38.869.113 I slot print_timing: id  0 | task 808 | 
+prompt eval time =     118.73 ms /    32 tokens (    3.71 ms per token,   269.51 tokens per second)
+       eval time =     211.94 ms /    14 tokens (   15.14 ms per token,    66.06 tokens per second)
+      total time =     330.67 ms /    46 tokens
+0.38.869.220 I slot      release: id  0 | task 808 | stop processing: n_tokens = 508, truncated = 0
+0.38.869.257 I srv  update_slots: all slots are idle
+0.38.890.188 I srv  params_from_: Chat format: peg-native
+0.38.890.530 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.967 (> 0.100 thold), f_keep = 0.807
+0.38.890.683 I reasoning-budget: activated, budget=2147483647 tokens
+0.38.890.717 I slot launch_slot_: id  0 | task 824 | processing task, is_child = 0
+0.38.890.724 W slot update_slots: id  0 | task 824 | n_past = 410, slot.prompt.tokens.size() = 508, seq_id = 0, pos_min = 507, n_swa = 0
+0.38.890.724 I slot update_slots: id  0 | task 824 | Checking checkpoint with [490, 490] against 410...
+0.38.890.725 I slot update_slots: id  0 | task 824 | Checking checkpoint with [419, 419] against 410...
+0.38.890.726 W slot update_slots: id  0 | task 824 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.38.890.729 W slot update_slots: id  0 | task 824 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.38.892.293 W slot update_slots: id  0 | task 824 | erased invalidated context checkpoint (pos_min = 490, pos_max = 490, n_tokens = 491, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.39.240.897 I slot create_check: id  0 | task 824 | created context checkpoint 1 of 32 (pos_min = 419, pos_max = 419, n_tokens = 420, size = 62.813 MiB)
+0.39.889.179 I slot print_timing: id  0 | task 824 | 
+prompt eval time =     380.09 ms /   424 tokens (    0.90 ms per token,  1115.53 tokens per second)
+       eval time =     618.35 ms /    40 tokens (   15.46 ms per token,    64.69 tokens per second)
+      total time =     998.43 ms /   464 tokens
+0.39.889.258 I slot      release: id  0 | task 824 | stop processing: n_tokens = 463, truncated = 0
+0.39.889.291 I srv  update_slots: all slots are idle
+0.39.916.715 I srv  params_from_: Chat format: peg-native
+0.39.917.343 I slot get_availabl: id  0 | task -1 | selected slot by LCP similarity, sim_best = 0.931 (> 0.100 thold), f_keep = 0.875
+0.39.917.545 I reasoning-budget: activated, budget=2147483647 tokens
+0.39.917.577 I slot launch_slot_: id  0 | task 866 | processing task, is_child = 0
+0.39.917.585 W slot update_slots: id  0 | task 866 | n_past = 405, slot.prompt.tokens.size() = 463, seq_id = 0, pos_min = 462, n_swa = 0
+0.39.917.586 I slot update_slots: id  0 | task 866 | Checking checkpoint with [419, 419] against 405...
+0.39.917.587 W slot update_slots: id  0 | task 866 | forcing full prompt re-processing due to lack of cache data (likely due to SWA or hybrid/recurrent memory, see https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
+0.39.917.590 W slot update_slots: id  0 | task 866 | erased invalidated context checkpoint (pos_min = 419, pos_max = 419, n_tokens = 420, n_swa = 0, pos_next = 0, size = 62.813 MiB)
+0.40.271.964 I slot create_check: id  0 | task 866 | created context checkpoint 1 of 32 (pos_min = 430, pos_max = 430, n_tokens = 431, size = 62.813 MiB)
+0.41.551.306 I slot print_timing: id  0 | task 866 | 
+prompt eval time =     384.12 ms /   435 tokens (    0.88 ms per token,  1132.46 tokens per second)
+       eval time =    1249.57 ms /    80 tokens (   15.62 ms per token,    64.02 tokens per second)
+      total time =    1633.69 ms /   515 tokens
+0.41.551.447 I slot      release: id  0 | task 866 | stop processing: n_tokens = 514, truncated = 0
+0.41.551.496 I srv  update_slots: all slots are idle
+0.41.552.857 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-vision-q106-faoff.log

@@ -0,0 +1,66 @@
+0.00.044.023 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.044.026 I device_info:
+0.00.044.069 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 125177 MiB free)
+0.00.044.147 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131926 MiB free)
+0.00.044.151 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.044.190 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.044.225 I srv          init: running without SSL
+0.00.044.247 I srv          init: using 31 threads for HTTP server
+0.00.044.248 I srv          init: the WebUI is disabled
+0.00.044.304 I srv         start: binding port with default address family
+0.00.045.454 I srv          main: loading model
+0.00.045.457 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.079.109 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.570.878 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.21.071.707 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.21.277.411 W load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
+0.21.277.414 W load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
+0.21.277.414 W load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842
+
+0.22.242.916 I srv    load_model: loaded multimodal model, '/mnt/models/nex-n2.5-mini/out/mmproj-Nex-N2.5-mini-BF16.gguf'
+0.22.242.925 I srv    load_model: initializing slots, n_slots = 1
+0.22.421.079 W srv    load_model: speculative decoding will use checkpoints
+0.22.421.086 W common_speculative_init: no implementations specified for speculative decoding
+0.22.421.088 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.22.421.122 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.22.421.129 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.22.421.144 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.22.429.869 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.22.436.195 I srv          init: init: chat template, thinking = 1
+0.22.436.220 I srv          main: model loaded
+0.22.436.223 I srv          main: server is listening on http://127.0.0.1:18600
+0.22.436.225 I srv  update_slots: all slots are idle
+0.24.206.034 I srv  params_from_: Chat format: peg-native
+0.24.206.169 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.206.172 I srv  get_availabl: updating prompt cache
+0.24.206.177 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.206.182 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.206.183 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.206.233 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.246.538 I srv  process_chun: processing image...
+0.24.375.032 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.375.099 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.571.922 I srv  process_chun: image processed in 326 ms
+0.24.572.141 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.572.170 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.650.730 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 34, pos_max = 34, n_tokens = 217, size = 62.813 MiB)
+0.25.135.589 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     471.32 ms /   221 tokens (    2.13 ms per token,   468.89 tokens per second)
+       eval time =     458.01 ms /    30 tokens (   15.27 ms per token,    65.50 tokens per second)
+      total time =     929.34 ms /   251 tokens
+0.25.135.616 I slot      release: id  0 | task 0 | stop processing: n_tokens = 250, truncated = 0
+0.25.135.619 I srv  update_slots: all slots are idle
+0.26.136.415 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-vision-q106-faon.log

@@ -0,0 +1,66 @@
+0.00.039.566 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.039.569 I device_info:
+0.00.039.614 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 125113 MiB free)
+0.00.039.689 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131926 MiB free)
+0.00.039.693 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.039.733 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.039.777 I srv          init: running without SSL
+0.00.039.798 I srv          init: using 31 threads for HTTP server
+0.00.039.799 I srv          init: the WebUI is disabled
+0.00.039.854 I srv         start: binding port with default address family
+0.00.041.005 I srv          main: loading model
+0.00.041.007 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.072.658 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.20.385.335 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.20.547.529 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.20.746.298 W load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
+0.20.746.300 W load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
+0.20.746.300 W load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842
+
+0.21.800.919 I srv    load_model: loaded multimodal model, '/mnt/models/nex-n2.5-mini/out/mmproj-Nex-N2.5-mini-BF16.gguf'
+0.21.800.929 I srv    load_model: initializing slots, n_slots = 1
+0.21.998.188 W srv    load_model: speculative decoding will use checkpoints
+0.21.998.194 W common_speculative_init: no implementations specified for speculative decoding
+0.21.998.195 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.21.998.237 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.21.998.246 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.21.998.259 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.22.007.086 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>'
+0.22.013.608 I srv          init: init: chat template, thinking = 1
+0.22.013.635 I srv          main: model loaded
+0.22.013.638 I srv          main: server is listening on http://127.0.0.1:18600
+0.22.013.640 I srv  update_slots: all slots are idle
+0.24.005.244 I srv  params_from_: Chat format: peg-native
+0.24.005.366 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.005.368 I srv  get_availabl: updating prompt cache
+0.24.005.373 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.005.376 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.005.377 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.005.401 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.032.329 I srv  process_chun: processing image...
+0.24.141.298 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.141.364 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.326.150 I srv  process_chun: image processed in 294 ms
+0.24.326.369 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.326.399 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.404.937 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 34, pos_max = 34, n_tokens = 217, size = 62.813 MiB)
+0.24.744.679 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     426.45 ms /   221 tokens (    1.93 ms per token,   518.23 tokens per second)
+       eval time =     312.81 ms /    21 tokens (   14.90 ms per token,    67.13 tokens per second)
+      total time =     739.26 ms /   242 tokens
+0.24.744.708 I slot      release: id  0 | task 0 | stop processing: n_tokens = 241, truncated = 0
+0.24.744.712 I srv  update_slots: all slots are idle
+0.25.745.484 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/b_n-vision-q106-roff-faon.log

@@ -0,0 +1,70 @@
+0.00.208.975 I log_info: verbosity = 3 (adjust with the `-lv N` CLI arg)
+0.00.208.993 I device_info:
+0.00.209.271 I   - ROCm0   : AMD Radeon Graphics (131072 MiB, 123872 MiB free)
+0.00.209.724 I   - Vulkan0 : AMD Radeon Graphics (RADV GFX1151) (132096 MiB, 131922 MiB free)
+0.00.209.744 I   - CPU     : AMD RYZEN AI MAX+ 395 w/ Radeon 8060S (127438 MiB, 127438 MiB free)
+0.00.209.942 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.00.210.045 I srv          init: running without SSL
+0.00.210.124 I srv          init: using 31 threads for HTTP server
+0.00.210.128 I srv          init: the WebUI is disabled
+0.00.210.347 I srv         start: binding port with default address family
+0.00.211.826 I srv          main: loading model
+0.00.211.839 I srv    load_model: loading model '/mnt/models/nex-n2.5-mini/out/Nex-N2.5-mini-Q4_0_ROCMFP4_STRIX_LEAN.gguf'
+0.00.341.319 W llama_model_loader: direct I/O is enabled, disabling mmap
+0.21.841.225 W llama_context: n_ctx_seq (65536) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.22.147.423 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.22.493.441 W load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
+0.22.493.444 W load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
+0.22.493.445 W load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842
+
+0.22.758.348 I srv    load_model: loaded multimodal model, '/mnt/models/nex-n2.5-mini/out/mmproj-Nex-N2.5-mini-BF16.gguf'
+0.22.758.359 I srv    load_model: initializing slots, n_slots = 1
+0.23.027.404 W srv    load_model: speculative decoding will use checkpoints
+0.23.027.422 W common_speculative_init: no implementations specified for speculative decoding
+0.23.027.426 I slot   load_model: id  0 | task -1 | new slot, n_ctx = 65536
+0.23.027.506 I srv    load_model: prompt cache RAM enabled: limit_mib=8192
+0.23.027.508 I srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
+0.23.027.604 I srv          init: idle slots will be saved to prompt cache upon starting a new task
+0.23.040.255 I init: chat template, example_format: '<|im_start|>system
+You are a helpful assistant<|im_end|>
+<|im_start|>user
+Hello<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+Hi there<|im_end|>
+<|im_start|>user
+How are you?<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+'
+0.23.049.208 I srv          init: init: chat template, thinking = 0
+0.23.049.258 I srv          main: model loaded
+0.23.049.261 I srv          main: server is listening on http://127.0.0.1:18600
+0.23.049.265 I srv  update_slots: all slots are idle
+0.24.037.221 I srv  params_from_: Chat format: peg-native
+0.24.037.451 I slot get_availabl: id  0 | task -1 | selected slot by LRU, t_last = -1
+0.24.037.454 I srv  get_availabl: updating prompt cache
+0.24.037.461 I srv          load:  - looking for better prompt, base f_keep = -1.000, sim = 0.000
+0.24.037.467 I srv        update:  - cache state: 0 prompts, 0.000 MiB (limits: 8192.000 MiB, 65536 tokens, 8589934592 est)
+0.24.037.470 I srv  get_availabl: prompt cache update took 0.01 ms
+0.24.037.553 I slot launch_slot_: id  0 | task 0 | processing task, is_child = 0
+0.24.096.831 I srv  process_chun: processing image...
+0.24.348.055 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.348.208 W find_slot: non-consecutive token position 4 after 3 for sequence 0 with 196 new tokens
+0.24.670.293 I srv  process_chun: image processed in 573 ms
+0.24.670.967 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.671.060 W find_slot: non-consecutive token position 34 after 4 for sequence 0 with 17 new tokens
+0.24.787.337 I slot create_check: id  0 | task 0 | created context checkpoint 1 of 32 (pos_min = 34, pos_max = 34, n_tokens = 217, size = 62.813 MiB)
+0.25.422.382 I slot print_timing: id  0 | task 0 | 
+prompt eval time =     788.44 ms /   221 tokens (    3.57 ms per token,   280.30 tokens per second)
+       eval time =     596.35 ms /    21 tokens (   28.40 ms per token,    35.21 tokens per second)
+      total time =    1384.78 ms /   242 tokens
+0.25.422.466 I slot      release: id  0 | task 0 | stop processing: n_tokens = 241, truncated = 0
+0.25.422.481 I srv  update_slots: all slots are idle
+0.26.423.773 I srv    operator(): operator(): cleaning up before exit...

recipe/logs/diag_bf16.log

@@ -0,0 +1,5 @@
+[2026-09-16T22:27:17Z] bf16_rocm_faoff rc=0 [1]136.6879 [2]166.5229 [3]174.2693 [4]169.2609 
+[2026-09-16T22:29:15Z] bf16_vk_faon rc=0 [1]5.6953 [2]6.6594 [3]7.0323 [4]7.2857 
+[2026-09-16T22:29:46Z] q106_vk_faon rc=0 [1]6.1880 [2]7.0357 [3]7.4624 [4]7.7345 
+[2026-09-16T22:30:26Z] bf16_cpu rc=0 [1]141.8336 [2]168.6793 
+[2026-09-16T22:30:26Z] DIAG_BF16_DONE

recipe/logs/diag_bf16_cpu.log

@@ -0,0 +1,13 @@
+0.00.032.646 I common_init_result: fitting params to device memory ...
+0.00.032.649 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.08.593.776 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.08.699.313 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.09.377.813 I 
+0.09.377.939 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.09.378.018 I perplexity: tokenizing the input ..
+0.09.680.733 I perplexity: tokenization took 302.776 ms
+0.09.680.810 I perplexity: calculating perplexity over 2 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.24.718.172 I perplexity: 15.04 seconds per pass - ETA 0.50 minutes
+[1]141.8336,[2]168.6793,
+0.39.305.358 I Final estimate: PPL = 168.6793 +/- 12.64504
+

recipe/logs/diag_bf16_purecpu_c1.log

@@ -0,0 +1,14 @@
+0.00.039.976 I common_init_result: fitting params to device memory ...
+0.00.039.980 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.00.307.048 I common_params_fit_impl: projected to use 66756 MiB of host memory vs. 127438 MiB of total host memory
+0.00.633.915 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.00.653.775 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.01.312.044 I 
+0.01.312.177 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.01.312.255 I perplexity: tokenizing the input ..
+0.01.611.737 I perplexity: tokenization took 299.542 ms
+0.01.611.820 I perplexity: calculating perplexity over 1 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.11.553.629 I perplexity: 9.94 seconds per pass - ETA 0.15 minutes
+[1]5.6964,
+0.11.592.689 I Final estimate: PPL = 5.6964 +/- 0.43869
+

recipe/logs/diag_bf16_rocm_faoff.log

@@ -0,0 +1,13 @@
+0.00.041.213 I common_init_result: fitting params to device memory ...
+0.00.041.216 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+2.25.152.472 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+2.25.329.776 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+2.26.102.739 I 
+2.26.102.951 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+2.26.102.958 I perplexity: tokenizing the input ..
+2.26.847.963 I perplexity: tokenization took 744.994 ms
+2.26.848.073 I perplexity: calculating perplexity over 4 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+2.32.003.043 I perplexity: 5.15 seconds per pass - ETA 0.33 minutes
+[1]136.6879,[2]166.5229,[3]174.2693,[4]169.2609,
+2.45.605.000 I Final estimate: PPL = 169.2609 +/- 8.97119
+

recipe/logs/diag_bf16_vk_faon.log

@@ -0,0 +1,13 @@
+0.00.081.666 I common_init_result: fitting params to device memory ...
+0.00.081.669 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+1.04.951.000 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+1.05.009.711 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+1.06.643.947 I 
+1.06.646.719 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+1.06.646.727 I perplexity: tokenizing the input ..
+1.07.413.989 I perplexity: tokenization took 767.253 ms
+1.07.414.080 I perplexity: calculating perplexity over 4 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+1.19.697.964 I perplexity: 12.28 seconds per pass - ETA 0.82 minutes
+[1]5.6953,[2]6.6594,[3]7.0323,[4]7.2857,
+1.57.415.169 I Final estimate: PPL = 7.2857 +/- 0.29353
+

recipe/logs/diag_ppl_q106_rocm_c4.log

@@ -0,0 +1,13 @@
+0.00.044.069 I common_init_result: fitting params to device memory ...
+0.00.044.071 I common_init_result: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)
+0.03.928.272 W llama_context: n_ctx_seq (2048) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
+0.03.976.550 W common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
+0.04.333.276 I 
+0.04.333.426 I system_info: n_threads = 16 (n_threads_batch = 16) / 32 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX_VNNI = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 | 
+0.04.333.434 I perplexity: tokenizing the input ..
+0.04.683.706 I perplexity: tokenization took 350.264 ms
+0.04.683.803 I perplexity: calculating perplexity over 4 chunks, n_ctx=2048, batch_size=2048, n_seq=1
+0.07.474.485 I perplexity: 2.79 seconds per pass - ETA 0.18 minutes
+[1]6.2788,[2]7.1110,[3]7.5138,[4]7.8414,
+0.14.413.643 I Final estimate: PPL = 7.8414 +/- 0.32420
+