qwen2.5-0.5b-expo-DPO-ES-1 / eval_results.json
hZzy's picture
End of training
b1020e6 verified
raw
history blame contribute delete
594 Bytes
{
"epoch": 2.69248937175248,
"eval_dpo_loss": 2.2591824531555176,
"eval_logits": -0.659694254398346,
"eval_logps": -83.41045379638672,
"eval_loss": 2.3420462608337402,
"eval_objective": 2.2591824531555176,
"eval_ranking_idealized": 0.5295031070709229,
"eval_ranking_idealized_expo": 0.5212215185165405,
"eval_ranking_simple": 0.5403726696968079,
"eval_regularize": 2.2591824531555176,
"eval_runtime": 308.2442,
"eval_samples": 5790,
"eval_samples_per_second": 18.784,
"eval_steps_per_second": 1.567,
"eval_wo_beta": 6.68360710144043
}