deepseek-8b-orpo-lora / all_results.json
zfz1's picture
End of training
1075d3a verified
raw
history blame
924 Bytes
{
"epoch": 1.9936102236421727,
"eval_log_odds_chosen": 0.05526316165924072,
"eval_log_odds_ratio": -0.6908599138259888,
"eval_logits/chosen": 30.44268798828125,
"eval_logits/rejected": 30.32880210876465,
"eval_logps/chosen": -0.7262152433395386,
"eval_logps/rejected": -0.7568790316581726,
"eval_loss": 1.1456295251846313,
"eval_nll_loss": 1.071852684020996,
"eval_rewards/accuracies": 0.5647059082984924,
"eval_rewards/chosen": -0.07262152433395386,
"eval_rewards/margins": 0.0030663711950182915,
"eval_rewards/rejected": -0.07568789273500443,
"eval_runtime": 247.9026,
"eval_samples": 5398,
"eval_samples_per_second": 21.775,
"eval_steps_per_second": 0.343,
"total_flos": 0.0,
"train_loss": 1.3676127867820935,
"train_runtime": 5524.9246,
"train_samples": 20000,
"train_samples_per_second": 7.24,
"train_steps_per_second": 0.056
}