llama-3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.3-20260428-045924/train_results.json

{
    "epoch": 0.9989528795811519,
    "total_flos": 0.0,
    "train_loss": 5.002585172153369,
    "train_runtime": 6534.5428,
    "train_samples": 61135,
    "train_samples_per_second": 9.356,
    "train_steps_per_second": 0.073
}