Qwen3-1.7B-MATH-GDPO/train_results.json

{
    "total_flos": 0.0,
    "train_loss": -0.09854654141236097,
    "train_runtime": 3102.2738,
    "train_samples": 1348,
    "train_samples_per_second": 0.869,
    "train_steps_per_second": 0.054
}