qwen3-8b-base-epsilon-dpo-ultrafeedback-4xh200-batch-128/train_results.json

{
    "epoch": 0.9989528795811519,
    "total_flos": 0.0,
    "train_loss": 5.1642030939865915,
    "train_runtime": 8287.5392,
    "train_samples": 61135,
    "train_samples_per_second": 7.377,
    "train_steps_per_second": 0.058
}