qwen3-8b-base-margin-dpo-ultrafeedback-4xh200-batch-128-20260423-040315/train_results.json

{
    "epoch": 0.9989528795811519,
    "total_flos": 0.0,
    "train_loss": 4.779813265150698,
    "train_runtime": 7822.2821,
    "train_samples": 61135,
    "train_samples_per_second": 7.815,
    "train_steps_per_second": 0.061
}