{
  "schema_version": 1,
  "method": "grpo",
  "model": {
    "name": "HuggingFaceTB/SmolLM2-135M-Instruct",
    "revision": "12fd25f77366fa6b3b4b768ec3050bf629380bac",
    "allow_download": false
  },
  "dataset": "../examples/training/prompts.jsonl",
  "output_dir": "../runs/reasoning-grpo",
  "seed": 42,
  "lora": {"r": 8, "alpha": 16, "dropout": 0.0, "target_modules": "all-linear"},
  "training": {
    "max_steps": 10,
    "max_completion_length": 128,
    "per_device_train_batch_size": 2,
    "gradient_accumulation_steps": 1,
    "num_generations": 2,
    "learning_rate": 0.000005,
    "temperature": 1.0,
    "beta": 0.0,
    "loss_type": "grpo",
    "bf16": false,
    "gradient_checkpointing": true
  }
}