mistral-sft-7b-dpo-qlora / eval_results.json
MichaelR207's picture
End of training
59f4671 verified
raw
history blame
561 Bytes
{
"epoch": 5.0,
"eval_logits/chosen": -2.9818739891052246,
"eval_logits/rejected": -3.045213222503662,
"eval_logps/chosen": -86.26663208007812,
"eval_logps/rejected": -122.861572265625,
"eval_loss": 0.6932176947593689,
"eval_rewards/accuracies": 0.5625,
"eval_rewards/chosen": 0.0022567124105989933,
"eval_rewards/margins": 0.000948949484154582,
"eval_rewards/rejected": 0.0013077628100290895,
"eval_runtime": 11.9466,
"eval_samples": 20,
"eval_samples_per_second": 1.674,
"eval_steps_per_second": 0.167
}