hf_quant_config.json

{
    "producer": {
        "name": "modelopt",
        "version": "0.34.1.dev141+g5db6ea47c"
    },
    "quantization": {
        "quant_algo": "NVFP4",
        "kv_cache_quant_algo": null,
        "group_size": 16,
        "exclude_modules": [
            "lm_head",
            "backbone.layers.0.mixer.in_proj",
            "backbone.layers.0.mixer.out_proj",
            "backbone.layers.1.mixer.up_proj",
            "backbone.layers.1.mixer.down_proj",
            "backbone.layers.54.mixer.in_proj",
            "backbone.layers.54.mixer.out_proj",
            "backbone.layers.55.mixer.up_proj",
            "backbone.layers.55.mixer.down_proj",
            "backbone.layers.0.mixer.conv1d",
            "backbone.layers.2.mixer.conv1d",
            "backbone.layers.4.mixer.conv1d",
            "backbone.layers.6.mixer.conv1d",
            "backbone.layers.7.mixer.conv1d",
            "backbone.layers.9.mixer.conv1d",
            "backbone.layers.11.mixer.conv1d",
            "backbone.layers.13.mixer.conv1d",
            "backbone.layers.14.mixer.q_proj",
            "backbone.layers.14.mixer.k_proj",
            "backbone.layers.14.mixer.v_proj",
            "backbone.layers.14.mixer.o_proj",
            "backbone.layers.16.mixer.conv1d",
            "backbone.layers.18.mixer.conv1d",
            "backbone.layers.20.mixer.conv1d",
            "backbone.layers.21.mixer.q_proj",
            "backbone.layers.21.mixer.k_proj",
            "backbone.layers.21.mixer.v_proj",
            "backbone.layers.21.mixer.o_proj",
            "backbone.layers.23.mixer.conv1d",
            "backbone.layers.25.mixer.conv1d",
            "backbone.layers.27.mixer.conv1d",
            "backbone.layers.29.mixer.conv1d",
            "backbone.layers.30.mixer.q_proj",
            "backbone.layers.30.mixer.k_proj",
            "backbone.layers.30.mixer.v_proj",
            "backbone.layers.30.mixer.o_proj",
            "backbone.layers.32.mixer.conv1d",
            "backbone.layers.34.mixer.conv1d",
            "backbone.layers.36.mixer.conv1d",
            "backbone.layers.38.mixer.conv1d",
            "backbone.layers.39.mixer.q_proj",
            "backbone.layers.39.mixer.k_proj",
            "backbone.layers.39.mixer.v_proj",
            "backbone.layers.39.mixer.o_proj",
            "backbone.layers.41.mixer.conv1d",
            "backbone.layers.43.mixer.conv1d",
            "backbone.layers.44.mixer.conv1d",
            "backbone.layers.46.mixer.conv1d",
            "backbone.layers.48.mixer.conv1d",
            "backbone.layers.50.mixer.conv1d",
            "backbone.layers.52.mixer.conv1d",
            "backbone.layers.54.mixer.conv1d"
        ]
    }
}