stablelm-2-1_6b-orpo-full-v1 / eval_results.json
vain05's picture
End of training
3e9e234 verified
raw
history blame contribute delete
630 Bytes
{
"epoch": 2.0,
"eval_log_odds_chosen": 0.1751251220703125,
"eval_log_odds_ratio": -0.679931640625,
"eval_logits/chosen": -2.484375,
"eval_logits/rejected": -2.46875,
"eval_logps/chosen": -1.2734375,
"eval_logps/rejected": -1.390625,
"eval_loss": 1.3789583444595337,
"eval_nll_loss": 1.3359375,
"eval_rewards/accuracies": 0.5625,
"eval_rewards/chosen": -0.1279296875,
"eval_rewards/margins": 0.01141357421875,
"eval_rewards/rejected": -0.138671875,
"eval_runtime": 28.0834,
"eval_samples": 750,
"eval_samples_per_second": 26.706,
"eval_steps_per_second": 1.139
}