sft/train_results.json

{
    "epoch": 0.9933774834437086,
    "total_flos": 1.0660613606866944e+16,
    "train_loss": 0.39357583204905194,
    "train_runtime": 105.2274,
    "train_samples_per_second": 2.87,
    "train_steps_per_second": 0.713
}