llama-3-8b-base-new-dpo-hh-harmless-4xh200-batch-64-q_t-0.5-s_star-0.85/train_results.json

{
    "epoch": 0.999244142101285,
    "total_flos": 0.0,
    "train_loss": 1.0861937751748378,
    "train_runtime": 1754.5635,
    "train_samples": 42336,
    "train_samples_per_second": 24.129,
    "train_steps_per_second": 0.377
}