Fix KLD precision (#325)

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-04-12 16:17:50 +02:00
committed by GitHub
parent a3b16affaf
commit 65fc77c285

View File

@@ -133,7 +133,7 @@ static double log_softmax(int n_vocab, const float * logits, uint16_t * log_prob
max_logit = std::max(max_logit, logits[i]);
min_logit = std::min(min_logit, logits[i]);
}
min_logit = std::max(min_logit, max_logit - 16);
min_logit = std::max(min_logit, max_logit - 24);
double sum_exp = 0.0;
for (int i = 0; i < n_vocab; ++i) {
sum_exp += expf(logits[i] - max_logit);