train_results.json

{
    "effective_tokens_per_sec": 1959.7027045726834,
    "epoch": 3.0,
    "total_flos": 2.73366675523371e+17,
    "train_loss": 0.06700010338739792,
    "train_runtime": 549.218,
    "train_samples_per_second": 40.82,
    "train_steps_per_second": 0.639
}