zephyr-dpo-qlora-gpt4-5e-7 / eval_results.json
just1nseo's picture
End of training
4c34425 verified
raw
history blame contribute delete
736 Bytes
{
"epoch": 1.0,
"eval_logits/chosen": -2.7608723640441895,
"eval_logits/rejected": -2.7231576442718506,
"eval_logps/chosen": -290.1167297363281,
"eval_logps/rejected": -266.18206787109375,
"eval_loss": 0.6861240267753601,
"eval_rewards/accuracies": 0.5910000205039978,
"eval_rewards/chosen": -0.05523300915956497,
"eval_rewards/margins": 0.020798461511731148,
"eval_rewards/margins_max": 0.1489465832710266,
"eval_rewards/margins_min": -0.09424488991498947,
"eval_rewards/margins_std": 0.08015545457601547,
"eval_rewards/rejected": -0.07603146880865097,
"eval_runtime": 427.9882,
"eval_samples": 2000,
"eval_samples_per_second": 4.673,
"eval_steps_per_second": 0.292
}