results/train_results.json

{
    "epoch": 3.0,
    "total_flos": 4.553211650587354e+16,
    "train_loss": 0.08937127710183461,
    "train_runtime": 4426.0374,
    "train_samples": 160000,
    "train_samples_per_second": 108.449,
    "train_steps_per_second": 1.695
}