{
  "model": "google/gemma-3-270m",
  "lora_config": {
    "r": 16,
    "lora_alpha": 32,
    "target_modules": [
      "qkv_proj",
      "gate_proj",
      "up_proj",
      "down_proj"
    ],
    "lora_dropout": 0.05,
    "bias": "none",
    "task_type": "CAUSAL_LM"
  },
  "training_config": {
    "num_epochs": 3,
    "per_device_train_batch_size": 1,
    "per_device_eval_batch_size": 1,
    "gradient_accumulation_steps": 32,
    "learning_rate": 0.0002,
    "lr_scheduler_type": "cosine",
    "warmup_ratio": 0.03,
    "weight_decay": 0.001,
    "max_grad_norm": 0.3,
    "optim": "paged_adamw_8bit"
  },
  "baseline_accuracy": 37.83,
  "target_accuracy": 68.0,
  "CRITICAL_FIX": "Proper label masking - only training on answer tokens",
  "format": "Question: <q>\\nAnswer: <a>"
}