{
  "stage": "train",
  "run_id": "qwen3_4b_dpo_v3",
  "config": {
    "method": "dpo",
    "base_model": "Qwen/Qwen3-4B-Instruct-2507",
    "data_paths": [
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/cleaned_jsonl/qa1.cleaned.jsonl",
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/cleaned_jsonl/report.cleaned.jsonl",
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/cleaned_jsonl/qa1.no_context.chosen.jsonl",
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/cleaned_jsonl/dpo_v3_extras.jsonl"
    ],
    "output_dir": "/data/gpfs/projects/punim2426/persuasion_runs/qwen3_4b_dpo_v3",
    "lr": 0.00001,
    "epochs": 5,
    "batch_size": 1,
    "grad_accum": 8,
    "max_len": 1024,
    "warmup_ratio": 0.05,
    "seed": 42,
    "optim": "adamw_torch",
    "checkpoint_steps": [
      216,
      648
    ],
    "lora_rank": 8,
    "lora_alpha": 16,
    "dpo_beta": 0.1,
    "dpo_rpo_alpha": 1.0,
    "rejected_paths": [
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/qa1_pos.jsonl",
      "/data/gpfs/projects/punim2426/Persuasion/Dataset/report_pos (1).jsonl"
    ]
  },
  "outputs": {
    "model_path": "/data/gpfs/projects/punim2426/persuasion_runs/qwen3_4b_dpo_v3/final_model"
  },
  "created_at": "2026-07-02T06:51:02+00:00"
}