{
  "bits": 4,
  "checkpoint_format": "gptq",
  "desc_act": false,
  "dynamic": {
    "-:.*embed_tokens.*": {},
    "-:.*lm_head.*": {},
    "-:.*norm.*": {},
    "-:.*vision.*": {},
    "-:.*visual.*": {}
  },
  "format": "gptq",
  "group_size": 128,
  "lm_head": false,
  "meta": {
    "act_group_aware": true,
    "activation_weighted_mse": true,
    "auto_forward_data_parallel": true,
    "calibration_dataset": "wikitext",
    "damp_auto_increment": 0.01,
    "damp_percent": 0.05,
    "dense_vram_strategy": "exclusive",
    "dense_vram_strategy_devices": null,
    "fallback": {
      "smooth": {
        "group_size_threshold": 128,
        "maxshrink": 0.9,
        "steps": 32,
        "type": "mse"
      },
      "strategy": "rtn",
      "threshold": "0.5%"
    },
    "foem": {
      "alpha": 0.25,
      "beta": 0.2,
      "device": "auto"
    },
    "gc_mode": "interval",
    "gptaq": null,
    "hessian": {
      "chunk_bytes": null,
      "chunk_size": null,
      "staging_dtype": "float32"
    },
    "mock_quantization": false,
    "moe_vram_strategy": "exclusive",
    "moe_vram_strategy_devices": null,
    "mse": 2.0,
    "mtp_patch": {
      "bits": 4,
      "foem_note": "Main model remains the base FOEM ns256 artifact; MTP tensors are quantized by direct MTP-aware GPTQ-Pro core capture.",
      "group_size": 128,
      "nsamples": 256,
      "quantized_key_count": 32,
      "quantized_modules": [
        "mtp.fc",
        "mtp.layers.0.self_attn.q_proj",
        "mtp.layers.0.self_attn.k_proj",
        "mtp.layers.0.self_attn.v_proj",
        "mtp.layers.0.self_attn.o_proj",
        "mtp.layers.0.mlp.gate_proj",
        "mtp.layers.0.mlp.up_proj",
        "mtp.layers.0.mlp.down_proj"
      ],
      "seqlen": 2048,
      "source_model": "Jackrong/Qwopus3.6-27B-Coder",
      "type": "mtp-aware-gptq-pro-core"
    },
    "nsamples": 256,
    "offload_to_disk": true,
    "offload_to_disk_path": "/workspace/disk_offload_qwopus36_27b_coder",
    "pack_impl": "cpu",
    "quantizer": [
      "gptqmodel:6.1.0-dev"
    ],
    "reference_recipe": "groxaxo/Qwen3.6-27B-GPTQ-Pro-4bit",
    "seqlen": 2048,
    "source_model": "Jackrong/Qwopus3.6-27B-Coder",
    "static_groups": false,
    "true_sequential": true,
    "uri": "https://github.com/modelcloud/gptqmodel",
    "wait_for_submodule_finalizers": false
  },
  "method": "gptq",
  "pack_dtype": "int32",
  "quant_method": "gptq",
  "sym": true
}