Files
latamgpt-1b-dpo-vanilla/train_results.json

20 lines
697 B
JSON
Raw Normal View History

{
"entropy": 1.178030303030303,
"epoch": 1.0,
"logits/chosen": -0.30206525592803823,
"logits/rejected": -0.0929858402560999,
"logps/chosen": -472.09090909090907,
"logps/rejected": -557.5757575757576,
"mean_token_accuracy": 0.7097648002884605,
"num_tokens": 358277178.0,
"rewards/accuracies": 0.7234848484848485,
"rewards/chosen": -0.24461780894886365,
"rewards/margins": 1.0697687322443181,
"rewards/rejected": -1.3149266098484849,
"total_flos": 4.975817267087409e+18,
"train_loss": 0.527936527665322,
"train_runtime": 22045.0555,
"train_samples": 272898,
"train_samples_per_second": 12.379,
"train_steps_per_second": 0.097
}