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

{
    "epoch": 5.0,
    "eval_accuracy": 0.9471856009625939,
    "eval_loss": 0.16564106941223145,
    "eval_runtime": 0.5723,
    "eval_samples_per_second": 8.737,
    "eval_steps_per_second": 3.495,
    "num_input_tokens_seen": 2242920
}