SenseNova-Loop-9B-NVFP4-MLP / hf_quant_config.json
Kaleto's picture
Quantize the LM head to NVFP4 (14 % faster decode, 1.5 GB smaller), add MTP and AWQ comparison to the card
ff00187 verified
Raw History Blame Contribute Delete
4.24 kB
{
"producer": {
"name": "modelopt",
"version": "0.43.0"
},
"quantization": {
"quant_algo": "W4A16_NVFP4",
"kv_cache_quant_algo": null,
"group_size": 16,
"exclude_modules": [
"model.language_model.layers.0.linear_attn*",
"model.language_model.layers.1.linear_attn*",
"model.language_model.layers.10.linear_attn*",
"model.language_model.layers.10.mhc_attn",
"model.language_model.layers.10.mhc_ffn",
"model.language_model.layers.11.mhc_attn",
"model.language_model.layers.11.mhc_ffn",
"model.language_model.layers.11.self_attn*",
"model.language_model.layers.12.linear_attn*",
"model.language_model.layers.12.mhc_attn",
"model.language_model.layers.12.mhc_ffn",
"model.language_model.layers.13.linear_attn*",
"model.language_model.layers.13.mhc_attn",
"model.language_model.layers.13.mhc_ffn",
"model.language_model.layers.14.linear_attn*",
"model.language_model.layers.14.mhc_attn",
"model.language_model.layers.14.mhc_ffn",
"model.language_model.layers.15.mhc_attn",
"model.language_model.layers.15.mhc_ffn",
"model.language_model.layers.15.self_attn*",
"model.language_model.layers.16.linear_attn*",
"model.language_model.layers.16.mhc_attn",
"model.language_model.layers.16.mhc_ffn",
"model.language_model.layers.17.linear_attn*",
"model.language_model.layers.17.mhc_attn",
"model.language_model.layers.17.mhc_ffn",
"model.language_model.layers.18.linear_attn*",
"model.language_model.layers.18.mhc_attn",
"model.language_model.layers.18.mhc_ffn",
"model.language_model.layers.19.mhc_attn",
"model.language_model.layers.19.mhc_ffn",
"model.language_model.layers.19.self_attn*",
"model.language_model.layers.2.linear_attn*",
"model.language_model.layers.20.linear_attn*",
"model.language_model.layers.20.mhc_attn",
"model.language_model.layers.20.mhc_ffn",
"model.language_model.layers.21.linear_attn*",
"model.language_model.layers.21.mhc_attn",
"model.language_model.layers.21.mhc_ffn",
"model.language_model.layers.22.linear_attn*",
"model.language_model.layers.22.mhc_attn",
"model.language_model.layers.22.mhc_ffn",
"model.language_model.layers.23.mhc_attn",
"model.language_model.layers.23.mhc_ffn",
"model.language_model.layers.23.self_attn*",
"model.language_model.layers.24.linear_attn*",
"model.language_model.layers.24.mhc_attn",
"model.language_model.layers.24.mhc_ffn",
"model.language_model.layers.25.linear_attn*",
"model.language_model.layers.25.mhc_attn",
"model.language_model.layers.25.mhc_ffn",
"model.language_model.layers.26.linear_attn*",
"model.language_model.layers.26.mhc_attn",
"model.language_model.layers.26.mhc_ffn",
"model.language_model.layers.27.mhc_attn",
"model.language_model.layers.27.mhc_ffn",
"model.language_model.layers.27.self_attn*",
"model.language_model.layers.28.linear_attn*",
"model.language_model.layers.29.linear_attn*",
"model.language_model.layers.3.self_attn*",
"model.language_model.layers.30.linear_attn*",
"model.language_model.layers.31.self_attn*",
"model.language_model.layers.4.linear_attn*",
"model.language_model.layers.4.mhc_attn",
"model.language_model.layers.4.mhc_ffn",
"model.language_model.layers.5.linear_attn*",
"model.language_model.layers.5.mhc_attn",
"model.language_model.layers.5.mhc_ffn",
"model.language_model.layers.6.linear_attn*",
"model.language_model.layers.6.mhc_attn",
"model.language_model.layers.6.mhc_ffn",
"model.language_model.layers.7.mhc_attn",
"model.language_model.layers.7.mhc_ffn",
"model.language_model.layers.7.self_attn*",
"model.language_model.layers.8.linear_attn*",
"model.language_model.layers.8.mhc_attn",
"model.language_model.layers.8.mhc_ffn",
"model.language_model.layers.9.linear_attn*",
"model.language_model.layers.9.mhc_attn",
"model.language_model.layers.9.mhc_ffn",
"model.visual*",
"mtp*"
]
}
}