OpenELM-1_1B-DPO / eval_results.json
CharlesLi's picture
End of training
ad3a23a verified
raw
history blame contribute delete
476 Bytes
{
"epoch": 3.0,
"eval_logits/chosen": -13.0,
"eval_logits/rejected": -13.0,
"eval_logps/chosen": -612.0,
"eval_logps/rejected": -632.0,
"eval_loss": 1.0008906126022339,
"eval_rewards/accuracies": 0.71875,
"eval_rewards/chosen": -14.5625,
"eval_rewards/margins": 2.609375,
"eval_rewards/rejected": -17.125,
"eval_runtime": 46.6328,
"eval_samples": 2000,
"eval_samples_per_second": 42.888,
"eval_steps_per_second": 0.686
}