llama3-8b-base-new-method-s_star0.6-20260425-180936/train_results.json

{
    "epoch": 0.9989528795811519,
    "total_flos": 0.0,
    "train_loss": 4.542374380479568,
    "train_runtime": 6039.2377,
    "train_samples": 61135,
    "train_samples_per_second": 10.123,
    "train_steps_per_second": 0.079
}