{
  "source_repo": "Qwen/Qwen3.8-27B",
  "source_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
  "forge_version": "2.7.1",
  "recipe_origin": "Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed",
  "body_bits": 4,
  "body_group_size": 32,
  "body_mode": "affine",
  "body_dtype": "bf16",
  "module_overrides": [
    {
      "bits": 8,
      "group_size": 64,
      "suffix": "embed_tokens"
    },
    {
      "bits": 8,
      "group_size": 64,
      "suffix": "lm_head"
    },
    {
      "bits": 8,
      "group_size": 64,
      "suffix": "linear_attn.out_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "layers": [
        56,
        57,
        58,
        59,
        60,
        61,
        62,
        63
      ],
      "suffix": "mlp.gate_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "layers": [
        56,
        57,
        58,
        59,
        60,
        61,
        62,
        63
      ],
      "suffix": "mlp.up_proj"
    },
    {
      "bits": 8,
      "group_size": 64,
      "layers": [
        56,
        57,
        58,
        59,
        60,
        61,
        62,
        63
      ],
      "suffix": "mlp.down_proj"
    }
  ],
  "mtp_policy": "keep_bf16"
}