SFT/train_results.json

{
    "epoch": 2.0,
    "total_flos": 2.6820605084455404e+18,
    "train_loss": 0.2813772309560665,
    "train_runtime": 15083.292,
    "train_samples_per_second": 4.374,
    "train_steps_per_second": 0.137
}