From c90e78edc3b87c9d3d91f3f1ef817561bac2a02f Mon Sep 17 00:00:00 2001 From: beiller Date: Sat, 11 Mar 2023 14:55:57 -0500 Subject: [PATCH] Numeric fix Should probably still scale by temp even if penalized --- utils.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/utils.cpp b/utils.cpp index 59fd05a..eceab45 100644 --- a/utils.cpp +++ b/utils.cpp @@ -392,7 +392,7 @@ gpt_vocab::id llama_sample_top_p( const double scale = 1.0/temp; for (int i = 0; i < n_logits; ++i) { if ( std::find(last_n_tokens.begin(), last_n_tokens.end(), i) != last_n_tokens.end() ) { - logits_id.push_back(std::make_pair(logits[i]*(1/repeat_penalty), i)); + logits_id.push_back(std::make_pair(logits[i]*scale*(1/repeat_penalty), i)); } else { logits_id.push_back(std::make_pair(logits[i]*scale, i)); }