qwen3-4B-instruct-refiner-sft/train_results.json

{
    "epoch": 5.0,
    "total_flos": 3.0080813400754176e+18,
    "train_loss": 0.12334943315905438,
    "train_runtime": 40705.595,
    "train_samples_per_second": 2.097,
    "train_steps_per_second": 0.066
}