qwen_uncCPO_entropy_0_01 / eval_results.json
yakazimir's picture
End of training
b7b361b verified
{
"epoch": 2.999297541394882,
"eval_logits/chosen": 0.05091719701886177,
"eval_logits/rejected": 0.14437255263328552,
"eval_logps/chosen": -4.325172424316406,
"eval_logps/rejected": -5.104434013366699,
"eval_loss": 0.04997371509671211,
"eval_rewards/accuracies": 0.68916916847229,
"eval_rewards/chosen": -4.325172424316406,
"eval_rewards/margins": 0.7792617678642273,
"eval_rewards/rejected": -5.104434013366699,
"eval_runtime": 44.0638,
"eval_samples": 1345,
"eval_samples_per_second": 30.524,
"eval_sft_loss": 3.922001361846924,
"eval_steps_per_second": 7.648
}