llama-3-8b-base-epsilon-dpo-hh-helpful-8xh200/train_results.json

{
    "epoch": 1.0,
    "total_flos": 0.0,
    "train_loss": 0.6232832217917723,
    "train_runtime": 1489.7896,
    "train_samples": 43598,
    "train_samples_per_second": 29.265,
    "train_steps_per_second": 0.228
}