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

{
    "epoch": 3.0,
    "total_flos": 62893478699008.0,
    "train_loss": 0.36721718521185326,
    "train_runtime": 28478.4394,
    "train_samples_per_second": 1.896,
    "train_steps_per_second": 0.007
}