zephyr-7b-dpo-qlora / eval_results.json
ale-bay's picture
End of training
fdadbc0 verified
raw
history blame
561 Bytes
{
"epoch": 1.0,
"eval_logits/chosen": 2.2971765995025635,
"eval_logits/rejected": 2.509765625,
"eval_logps/chosen": -516.2817993164062,
"eval_logps/rejected": -600.310302734375,
"eval_loss": 0.4919675290584564,
"eval_rewards/accuracies": 0.7559999823570251,
"eval_rewards/chosen": -2.5097556114196777,
"eval_rewards/margins": 1.0807288885116577,
"eval_rewards/rejected": -3.590484380722046,
"eval_runtime": 449.3054,
"eval_samples": 2000,
"eval_samples_per_second": 4.451,
"eval_steps_per_second": 0.278
}