Qwen3-8B-SFT-envbench_qwen-all/train_results.json

{
    "epoch": 5.0,
    "num_input_tokens_seen": 36600520,
    "total_flos": 1.6620454705385964e+18,
    "train_loss": 0.11312562568641421,
    "train_runtime": 4017.5693,
    "train_samples_per_second": 1.325,
    "train_steps_per_second": 0.083
}