TurboQuant weight quantization: 4-bit, group_size=128
Model: /home/ckl/.cache/modelscope/hub/Qwen/Qwen3___5-9B
Output: /home/ckl/.cache/modelscope/hub/Qwen/Qwen3___5-9B-TQ4
  Copied configuration.json
  Copied config.json
  Copied merges.txt
  Copied preprocessor_config.json
  Copied model.safetensors.index.json
  Copied video_preprocessor_config.json
  Copied tokenizer_config.json
  Copied vocab.json
  Copied tokenizer.json
Computing Lloyd-Max codebook for D=128, bits=4...
  Centroids: [-0.23777656 -0.18096589 -0.14193912 -0.11041539 -0.08293881 -0.05785765
 -0.0342055  -0.01132094  0.01132094  0.0342055   0.05785765  0.08293881
  0.11041539  0.14193912  0.18096589  0.23777656]

Processing model.safetensors-00001-of-00004.safetensors...
  model.language_model.layers.14.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.14.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.14.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.15.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.15.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.15.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.25.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.26.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.26.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.26.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.27.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.27.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  Saved model.safetensors-00001-of-00004.safetensors

Processing model.safetensors-00002-of-00004.safetensors...
  model.language_model.layers.0.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.0.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.0.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.1.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.1.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.1.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.10.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.10.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.10.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.11.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.11.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.11.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.16.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.16.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.16.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.17.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.17.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.17.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.18.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.18.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.18.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.19.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.21.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.21.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.21.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.22.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.22.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.22.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.23.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.23.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.23.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.27.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.3.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.3.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.3.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.30.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.30.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.30.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.31.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.31.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.31.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.4.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.4.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.4.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.8.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.8.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.8.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.9.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.9.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.9.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  mtp.layers.0.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  mtp.layers.0.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  mtp.layers.0.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  Saved model.safetensors-00002-of-00004.safetensors

Processing model.safetensors-00003-of-00004.safetensors...
  model.language_model.layers.11.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.12.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.12.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.12.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.13.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.13.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.13.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.15.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.19.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.19.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.19.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.2.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.2.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.2.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.20.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.20.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.20.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.23.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.24.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.24.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.24.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.25.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.25.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.27.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.28.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.28.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.28.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.29.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.29.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.29.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.3.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.31.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  model.language_model.layers.5.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.5.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.5.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.6.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.6.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.6.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.7.mlp.down_proj.weight: [4096, 12288] → packed=(4096, 6144) scales=(4096, 96) (3.9x compression)
  model.language_model.layers.7.mlp.gate_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.7.mlp.up_proj.weight: [12288, 4096] → packed=(12288, 2048) scales=(12288, 32) (3.9x compression)
  model.language_model.layers.7.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  mtp.layers.0.self_attn.q_proj.weight: [8192, 4096] → packed=(8192, 2048) scales=(8192, 32) (3.9x compression)
  Saved model.safetensors-00003-of-00004.safetensors

Processing model.safetensors-00004-of-00004.safetensors...
  model.language_model.layers.11.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.11.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.11.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.15.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.15.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.15.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.19.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.19.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.19.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.23.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.23.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.23.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.27.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.27.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.27.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.3.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.3.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.3.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.31.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.31.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.31.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.7.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  model.language_model.layers.7.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  model.language_model.layers.7.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  mtp.layers.0.self_attn.k_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  mtp.layers.0.self_attn.o_proj.weight: [4096, 4096] → packed=(4096, 2048) scales=(4096, 32) (3.9x compression)
  mtp.layers.0.self_attn.v_proj.weight: [1024, 4096] → packed=(1024, 2048) scales=(1024, 32) (3.9x compression)
  Saved model.safetensors-00004-of-00004.safetensors

Updated model.safetensors.index.json (1045 entries)

Total: 11.02 GB → 2.84 GB (3.9x compression)
Config: /home/ckl/.cache/modelscope/hub/Qwen/Qwen3___5-9B-TQ4/turboquant_config.json
