qwen2.5-3b-instruct/train_results.json

{
    "epoch": 3.0,
    "total_flos": 5.2362750737986355e+17,
    "train_loss": 0.4823716257659482,
    "train_runtime": 11702.1548,
    "train_samples_per_second": 4.615,
    "train_steps_per_second": 0.018
}