sft/train_results.json

{
    "epoch": 4.998929718159115,
    "total_flos": 9.494243434291003e+18,
    "train_loss": 0.3732268425964336,
    "train_runtime": 53791.2525,
    "train_samples_per_second": 10.422,
    "train_steps_per_second": 0.326
}