Qwen2.5-3B-ha_grpo/train_results.json

{
    "total_flos": 0.0,
    "train_loss": 0.013747031251953628,
    "train_runtime": 124161.3091,
    "train_samples": 7498,
    "train_samples_per_second": 0.06,
    "train_steps_per_second": 0.002
}