stage1_30784a0a20/stage1_config.json

{
  "base_model": "Qwen/Qwen2.5-1.5B-Instruct",
  "batch_size": 16,
  "evidence_placeholder": false,
  "grad_accum": 1,
  "learning_rate": 2e-05,
  "lora_alpha": 16,
  "lora_r": 16,
  "max_eval_samples": 64,
  "max_seq_length": 1024,
  "max_train_samples": null,
  "num_epochs": 1,
  "save_mode": "adapter"
}