QLoRA_french_dpo/train_results.json

{
    "epoch": 1.0,
    "total_flos": 5.747405857487585e+17,
    "train_loss": 0.08356396047416255,
    "train_runtime": 17143.627,
    "train_samples_per_second": 0.583,
    "train_steps_per_second": 0.583
}