From c7c79b16cd34cc47d9a49f4bcdd1c535131ab7b3 Mon Sep 17 00:00:00 2001
From: Shi Shuai <126407087+shuaills@users.noreply.github.com>
Date: Fri, 21 Feb 2025 17:24:15 +0000
Subject: [PATCH] [Fix] OpenAI API adapter tokenizer encoding (#3432)

---
 python/sglang/srt/openai_api/adapter.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/python/sglang/srt/openai_api/adapter.py b/python/sglang/srt/openai_api/adapter.py
index 8972d42bc..0556f852a 100644
--- a/python/sglang/srt/openai_api/adapter.py
+++ b/python/sglang/srt/openai_api/adapter.py
@@ -941,7 +941,13 @@ def v1_chat_generate_request(
                     )
 
                 if assistant_prefix:
-                    prompt_ids += tokenizer_manager.tokenizer.encode(assistant_prefix)
+                    encoded = tokenizer_manager.tokenizer.encode(assistant_prefix)
+                    if (
+                        encoded
+                        and encoded[0] == tokenizer_manager.tokenizer.bos_token_id
+                    ):
+                        encoded = encoded[1:]
+                    prompt_ids += encoded
                 stop = request.stop
                 image_data = None
                 modalities = []