zephyr-7b-dpo-qlora / eval_results.json
shichengshuai98's picture
End of training
8167608 verified
{
"epoch": 0.9999018549416037,
"eval_logits/chosen": -1.3436298370361328,
"eval_logits/rejected": -1.2241628170013428,
"eval_logps/chosen": -593.8455200195312,
"eval_logps/rejected": -712.9634399414062,
"eval_loss": 0.4840041399002075,
"eval_rewards/accuracies": 0.7570000290870667,
"eval_rewards/chosen": -3.2860074043273926,
"eval_rewards/margins": 1.4316543340682983,
"eval_rewards/rejected": -4.717660903930664,
"eval_runtime": 695.2646,
"eval_samples": 2000,
"eval_samples_per_second": 2.877,
"eval_steps_per_second": 0.18
}