llama3-1B-sft/train_results.json

{
    "epoch": 3.0,
    "total_flos": 2.043435500286509e+18,
    "train_loss": 0.016654981696585226,
    "train_runtime": 5294.7714,
    "train_samples_per_second": 9.403,
    "train_steps_per_second": 0.294
}