qwen3-8b-base-epsilon-dpo-hh-helpful-4xh200-batch-64-20260424-040306/train_results.json

{
    "epoch": 1.0,
    "total_flos": 0.0,
    "train_loss": 1.0212064078200755,
    "train_runtime": 3308.3388,
    "train_samples": 43598,
    "train_samples_per_second": 13.178,
    "train_steps_per_second": 0.206
}