NuminaMath-7B-TIR/train_results.json

{
    "epoch": 4.0,
    "total_flos": 340525408321536.0,
    "train_loss": 0.3252074327502578,
    "train_runtime": 5343.4474,
    "train_samples": 73061,
    "train_samples_per_second": 41.505,
    "train_steps_per_second": 1.297
}