From: slaren <redacted>
Date: Sun, 8 Sep 2024 10:41:51 +0000 (+0200)
Subject: llama : sanitize tokens in the upper bound (#9359)
X-Git-Tag: upstream/0.0.4488~786
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=eae597182cb61bbde0b26e7cec5999d28b9327fe;p=pkg%2Fggml%2Fsources%2Fllama.cpp

llama : sanitize tokens in the upper bound (#9359)
---

diff --git a/src/llama.cpp b/src/llama.cpp
index f590bcd3..39e20440 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -16077,7 +16077,7 @@ static int llama_decode_internal(
     }
 
     for (uint32_t i = 0; i < n_tokens_all; ++i) {
-        if (batch_all.token[i] < 0) {
+        if (batch_all.token[i] < 0 || (uint32_t)batch_all.token[i] >= lctx.model.vocab.n_vocab) {
             LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch_all.token[i]);
             return -1;
         }
@@ -16376,7 +16376,7 @@ static int llama_encode_internal(
     }
 
     for (uint32_t i = 0; i < n_tokens; ++i) {
-        if (batch.token[i] < 0) {
+        if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= lctx.model.vocab.n_vocab) {
             LLAMA_LOG_ERROR("%s: invalid token[%d] = %d", __func__, i, batch.token[i]);
             return -1;
         }