From 8960fe86ae075c846c5df8848230d1904ba8877f Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Mon, 22 Apr 2024 15:41:11 +0300
Subject: [PATCH] llama : fix typo in <|im_end|> token text (#6745)

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

diff --git a/llama.cpp b/llama.cpp
index 7440c740f..a25d115c1 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -4340,7 +4340,7 @@ static void llm_load_vocab(
             }
         }
 
-        // find EOT token: "<|eot_id|>", "<|im_emd|>", "<end_of_turn>", etc.
+        // find EOT token: "<|eot_id|>", "<|im_end|>", "<end_of_turn>", etc.
         //
         // TODO: convert scripts should provide this token through the KV metadata LLAMA_KV_TOKENIZER_EOT_ID
         //       for now, we apply this workaround to find the EOT token based on its text
@@ -4351,7 +4351,7 @@ static void llm_load_vocab(
                         //       need to fix convert script
                         //vocab.id_to_token[t.second].type == LLAMA_TOKEN_TYPE_CONTROL &&
                         (t.first == "<|eot_id|>" ||
-                         t.first == "<|im_emd|>" ||
+                         t.first == "<|im_end|>" ||
                          t.first == "<end_of_turn>"
                         )
                    ) {