llama-3-8b-base-margin-dpo-4xh100-real/train_results.json

{
    "epoch": 0.9988221436984688,
    "total_flos": 0.0,
    "train_loss": 0.5733983517942689,
    "train_runtime": 5703.1485,
    "train_samples": 61135,
    "train_samples_per_second": 10.72,
    "train_steps_per_second": 0.084
}