From 711ef84442117b07c345b57fa16ceb5667b0dd03 Mon Sep 17 00:00:00 2001
From: Arianna Method <theariannamethod@gmail.com>
Date: Wed, 23 Sep 2026 10:45:06 +0300
Subject: [PATCH] common : fix for two functions when top_k exceeds the
 vocabulary size. (ggml/1633)

---
 examples/common.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/examples/common.cpp b/examples/common.cpp
index 8eb633e58..1ff8b43ab 100644
--- a/examples/common.cpp
+++ b/examples/common.cpp
@@ -406,6 +406,7 @@ gpt_vocab::id gpt_sample_top_k_top_p(
         double temp,
         std::mt19937 & rng) {
     int n_logits = vocab.id_to_token.size();
+    top_k = std::min(top_k, n_logits);
 
     std::vector<std::pair<double, gpt_vocab::id>> logits_id;
     logits_id.reserve(n_logits);
@@ -491,6 +492,7 @@ gpt_vocab::id gpt_sample_top_k_top_p_repeat(
         std::mt19937 & rng) {
 
     int n_logits = vocab.id_to_token.size();
+    top_k = std::min(top_k, n_logits);
 
     const auto * plogits = logits;