NuminaMath-7B-CoT/train_results.json

{
    "epoch": 3.0,
    "total_flos": 2049636776804352.0,
    "train_loss": 0.42822988295175696,
    "train_runtime": 30718.8165,
    "train_samples": 863474,
    "train_samples_per_second": 21.729,
    "train_steps_per_second": 0.679
}