Meta-Llama-3-70B-Instruct / eval_results.json
statking's picture
End of training
89846e1 verified
{
"epoch": 0.9999296814570002,
"eval_log_odds_chosen": 0.3767525255680084,
"eval_log_odds_ratio": -0.62775719165802,
"eval_logits/chosen": 0.20116207003593445,
"eval_logits/rejected": -0.0032512724865227938,
"eval_logps/chosen": -0.8883616924285889,
"eval_logps/rejected": -1.1382248401641846,
"eval_loss": 1.2883824110031128,
"eval_nll_loss": 1.2074565887451172,
"eval_rewards/accuracies": 0.6132478713989258,
"eval_rewards/chosen": -0.08883616328239441,
"eval_rewards/margins": 0.02498631551861763,
"eval_rewards/rejected": -0.11382248252630234,
"eval_runtime": 975.3638,
"eval_samples": 1869,
"eval_samples_per_second": 1.916,
"eval_steps_per_second": 0.48
}