llama3-3B-sft/train_results.json

{
    "epoch": 3.0,
    "total_flos": 5.918873234148688e+18,
    "train_loss": 0.010715998370104875,
    "train_runtime": 13024.8275,
    "train_samples_per_second": 3.822,
    "train_steps_per_second": 0.06
}