zephyr-7b-dpo-full / eval_results.json
happii's picture
End of training
7ed7104 verified
raw
history blame
583 Bytes
{
"epoch": 2.998430141287284,
"eval_logits/chosen": -1.6524701118469238,
"eval_logits/rejected": -1.3897571563720703,
"eval_logps/chosen": -346.02056884765625,
"eval_logps/rejected": -366.2360534667969,
"eval_loss": 0.910855770111084,
"eval_rewards/accuracies": 0.7658730149269104,
"eval_rewards/chosen": -6.40665864944458,
"eval_rewards/margins": 4.195058822631836,
"eval_rewards/rejected": -10.601717948913574,
"eval_runtime": 196.377,
"eval_samples": 2000,
"eval_samples_per_second": 10.184,
"eval_steps_per_second": 0.321
}