{
  "stage": "s1",
  "base_model": "Qwen/Qwen3.5-0.8B",
  "resume_from": "jeopardy_g4_lora",
  "data": "training_data/board_train.jsonl",
  "lr": 5e-05,
  "epochs": 1.0,
  "max_steps": null,
  "batch_size": 2,
  "grad_accum": 8,
  "lora_r": 16,
  "final_loss": 0.14736651238941012,
  "global_step": 189
}