mistral-7b-base-epsilon-dpo-hh-helpful-4xh200-batch-64/train_results.json

{
    "epoch": 1.0,
    "total_flos": 0.0,
    "train_loss": 0.6833117403385398,
    "train_runtime": 3053.5655,
    "train_samples": 43598,
    "train_samples_per_second": 14.278,
    "train_steps_per_second": 0.223
}