Qwen3-8B-SFT-envbench_gpt5-yellow-green/train_results.json

{
    "epoch": 5.0,
    "num_input_tokens_seen": 31362728,
    "total_flos": 1.4241950524474655e+18,
    "train_loss": 0.4095978515633082,
    "train_runtime": 3265.849,
    "train_samples_per_second": 1.442,
    "train_steps_per_second": 0.09
}