training_info.json

{
  "version": "v2_updated_datasets",
  "strategy": "smollm3_3b_lora_hard_merge",
  "final_loss": 0.3239577183436989,
  "config": {
    "model_path": "HuggingFaceTB/SmolLM3-3B",
    "output_path": "./fyodor-mini",
    "use_auth_token": true,
    "batch_size": 8,
    "gradient_accumulation_steps": 8,
    "learning_rate": 0.0002,
    "num_epochs": 3,
    "warmup_steps": 100,
    "max_seq_length": 1024,
    "mixed_precision": true,
    "dtype": "bfloat16"
  }
}