llama3.1-cpo_j-full-0911 / eval_results.json
jbjeong91's picture
End of training
3047c65 verified
raw
history blame contribute delete
627 Bytes
{
"epoch": 4.9927766541462,
"eval_logits/chosen": -0.14757364988327026,
"eval_logits/rejected": -0.11364421248435974,
"eval_logps/chosen": -141.49264526367188,
"eval_logps/rejected": -155.7095184326172,
"eval_loss": 1.4373149871826172,
"eval_nll_loss": 0.17254449427127838,
"eval_rewards/accuracies": 0.654347836971283,
"eval_rewards/chosen": -14.149263381958008,
"eval_rewards/margins": 1.4216874837875366,
"eval_rewards/rejected": -15.570951461791992,
"eval_runtime": 76.603,
"eval_samples": 1826,
"eval_samples_per_second": 23.837,
"eval_steps_per_second": 1.501
}