qwen2.5-7b-instruct/train_results.json

{
    "epoch": 3.0,
    "total_flos": 1.0308479847759872e+17,
    "train_loss": 1.8930653027245696,
    "train_runtime": 5673.0769,
    "train_samples_per_second": 9.519,
    "train_steps_per_second": 0.038
}