{
  "format_version": 2,
  "base_model": "Qwen/Qwen3.5-4B-Base",
  "base_revision": "1001bb4d826a52d1f399e183466143f4da7b741b",
  "head": "bilinear",
  "head_width": 512,
  "head_last_token": true,
  "brier_weight": 1.0,
  "max_length": 4096,
  "dependencies": {
    "torch": "2.11.0+cu130",
    "transformers": "5.17.0",
    "peft": "0.21.0",
    "accelerate": "1.15.0",
    "bitsandbytes": "0.50.2",
    "numpy": "2.1.3"
  },
  "temperatures": {
    "choice": 1.681792830507429,
    "score": 1.4142135623730951,
    "noul": 1.5422108254079407
  },
  "images": {
    "max_image_tokens": 256,
    "placeholder": "<|image|>"
  },
  "temperature_calibration": {
    "fitted_on": "dev questions with a checked answer (hard sets, definitions), as that answer",
    "fit_examples_per_primitive": 1000,
    "teacher_data": "/content/teacher-data/training-data-clean-hard-v2, /content/teacher-data/training-data-clean-hard-v1, /content/teacher-data/training-data-clean-breadth-v1",
    "exported_temperatures": {
      "choice": 1.0,
      "score": 1.0,
      "noul": 1.0
    },
    "quantize4bit": false,
    "tied": [],
    "one_hot": false
  },
  "checkpoint": {
    "path": "/content/drive/MyDrive/jev-training/yev-qwen3.5-4b-clean-breadth/checkpoints/last.pt",
    "step": null
  },
  "soup": {
    "method": "exact weighted sum of LoRA weight changes (stacked to rank 128); heads averaged",
    "ingredients": [
      {
        "export": "/content/soups/breadth",
        "weight": 0.3,
        "init_checkpoint": null,
        "selected_by": null
      },
      {
        "export": "/content/drive/MyDrive/v2/yev-qwen3.5-4b-clean-end-to-end/exports/sharp-0.5-verified",
        "weight": 0.7,
        "init_checkpoint": null,
        "selected_by": null
      }
    ]
  }
}