From 124c718c73f915f3e4235ae2af8841356e76177d Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Tue, 13 Dec 2022 23:13:55 +0200
Subject: [PATCH] whisper : fix UB when reading buffer of length 0 bytes (#265)

---
 whisper.cpp | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/whisper.cpp b/whisper.cpp
index 33f258bc..1bc79967 100644
--- a/whisper.cpp
+++ b/whisper.cpp
@@ -549,13 +549,20 @@ static bool whisper_model_load(const std::string & fname, whisper_context & wctx
         //}
 
         std::string word;
+        std::vector<char> tmp;
         for (int i = 0; i < n_vocab; i++) {
             uint32_t len;
             read_safe(fin, len);
 
-            std::vector<char> tmp(len); // create a buffer
-            fin.read( &tmp[0], tmp.size() ); // read to buffer
-            word.assign(&tmp[0], tmp.size());
+            if (len > 0) {
+                tmp.resize(len);
+                fin.read(&tmp[0], tmp.size()); // read to buffer
+                word.assign(&tmp[0], tmp.size());
+            } else {
+                // seems like we have an empty-string token in multi-language models (i = 50256)
+                //fprintf(stderr, "%s: warning: empty-string token in vocab, i = %d\n", __func__, i);
+                word = "";
+            }
 
             vocab.token_to_id[word] = i;
             vocab.id_to_token[i] = word;