Llama3.2-3B_Paper_Impact_SFT/train_results.json

{
    "epoch": 3.0,
    "total_flos": 6.274047864041636e+18,
    "train_loss": 0.03664169063040653,
    "train_runtime": 42810.6565,
    "train_samples_per_second": 6.203,
    "train_steps_per_second": 0.048
}