qwen3-4b-base-prompt/train_results.json

{
    "epoch": 3.0,
    "total_flos": 5.550091697158881e+18,
    "train_loss": 0.008026252662842327,
    "train_runtime": 22347.8789,
    "train_samples_per_second": 2.327,
    "train_steps_per_second": 0.073
}