hf_quant_config.json

{
  "producer": {
    "name": "modelopt",
    "version": "0.42.0"
  },
  "quantization": {
    "quant_algo": "NVFP4",
    "group_size": 16,
    "kv_cache_quant_algo": "FP8",
    "exclude_modules": [
      "*.block_sparse_moe.gate",
      "lm_head",
      "embed_tokens",
      "norm",
      "mtp_modules",
      "e_score_correction_bias"
    ]
  }
}