zephyr-7b-dpo-full / eval_results.json
BubbleJoe's picture
End of training
aeda3e7 verified
{
"epoch": 1.0,
"eval_logits/chosen": 0.9002982378005981,
"eval_logits/rejected": 1.7371047735214233,
"eval_logps/chosen": -352.4257507324219,
"eval_logps/rejected": -450.12908935546875,
"eval_loss": 0.5019991993904114,
"eval_rewards/accuracies": 0.78125,
"eval_rewards/chosen": -0.8985068202018738,
"eval_rewards/margins": 0.9759426116943359,
"eval_rewards/rejected": -1.8744494915008545,
"eval_runtime": 39.9715,
"eval_samples": 2000,
"eval_samples_per_second": 50.036,
"eval_steps_per_second": 0.801
}