llama32-3b-redo-dpo_simple/train_results.json

{
    "epoch": 1.0,
    "total_flos": 1.761935585720664e+18,
    "train_loss": 0.5203882457111775,
    "train_runtime": 14354.8151,
    "train_samples_per_second": 4.13,
    "train_steps_per_second": 0.065
}