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

{
    "epoch": 5.0,
    "eval_accuracy": 0.8568963869568719,
    "eval_loss": 0.4850449860095978,
    "eval_runtime": 4.2558,
    "eval_samples_per_second": 11.749,
    "eval_steps_per_second": 3.055,
    "num_input_tokens_seen": 31362728
}