llama-3-8b-base-margin-dpo-hh-harmless-beta0.01/train_results.json

{
    "epoch": 0.999244142101285,
    "total_flos": 0.0,
    "train_loss": 1.1443162902220294,
    "train_runtime": 1702.0491,
    "train_samples": 42336,
    "train_samples_per_second": 24.874,
    "train_steps_per_second": 0.388
}