zephyr-7b-dpo-lora-r16 / eval_results.json
LaoRay's picture
End of training
3db1ae4 verified
raw
history blame contribute delete
569 Bytes
{
"epoch": 19.84,
"eval_logits/chosen": -2.154245376586914,
"eval_logits/rejected": -2.1524815559387207,
"eval_logps/chosen": -489.4880065917969,
"eval_logps/rejected": -554.3574829101562,
"eval_loss": 0.6410359144210815,
"eval_rewards/accuracies": 0.6650000214576721,
"eval_rewards/chosen": -2.2125351428985596,
"eval_rewards/margins": 0.8465680480003357,
"eval_rewards/rejected": -3.059103488922119,
"eval_runtime": 66.398,
"eval_samples": 200,
"eval_samples_per_second": 3.012,
"eval_steps_per_second": 0.377
}