From 8de3d5aa50bf4c8a444813430bedb6abc21f92f8 Mon Sep 17 00:00:00 2001
From: Nexesenex <124105151+Nexesenex@users.noreply.github.com>
Date: Mon, 12 Aug 2024 22:34:09 +0200
Subject: [PATCH] preclamp to the first 100 logits

---
 gpttype_adapter.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 53b7f789fa37cc..4479f34f76b42d 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -843,8 +843,8 @@ int mirostat, float mirostat_tau, float mirostat_eta, float dry_multiplier, floa
         sample_grammar(file_format, n_vocab, &candidates_p, grammar);
     }
 
-    //prefilter to top 5k tokens for improved speed
-    llama_sample_top_k(nullptr, &candidates_p, 5000, 1);
+    //prefilter to top "pick a nucmber" tokens for improved speed (KCPP-official default is 5k)
+    llama_sample_top_k(nullptr, &candidates_p, 100, 1);
 
     if (mirostat == 1 || mirostat == 2)
     {