qwen_cpo_entropy_0_3 / eval_results.json
yakazimir's picture
End of training
d6c4c57 verified
raw
history blame contribute delete
628 Bytes
{
"epoch": 2.999297541394882,
"eval_logits/chosen": 0.1268635243177414,
"eval_logits/rejected": 0.2186587154865265,
"eval_logps/chosen": -1.3990017175674438,
"eval_logps/rejected": -1.8440245389938354,
"eval_loss": 1.0415802001953125,
"eval_rewards/accuracies": 0.6157270073890686,
"eval_rewards/chosen": -1.3990017175674438,
"eval_rewards/margins": 0.44502273201942444,
"eval_rewards/rejected": -1.8440245389938354,
"eval_runtime": 49.8615,
"eval_samples": 1345,
"eval_samples_per_second": 26.975,
"eval_sft_loss": 1.4031308889389038,
"eval_steps_per_second": 6.759
}