From b399d94de3edd94c0530f17d9ad00c48236c7b62 Mon Sep 17 00:00:00 2001
From: ViperEkura <3081035982@qq.com>
Date: Mon, 15 Jun 2026 14:05:31 +0800
Subject: [PATCH] fix: add tokenizer thread-safety patch and disable frontend
 multiprocessing for image support

---
 server.py | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/server.py b/server.py
index 7d3e6c3..818826a 100644
--- a/server.py
+++ b/server.py
@@ -1,8 +1,22 @@
 import os, sys
 import asyncio
 import subprocess
+import threading
 from pathlib import Path
 
+import transformers
+from transformers import AddedToken, PreTrainedTokenizerBase
+
+# ── monkey-patch: make tokenizer encode thread-safe ──
+_tokenizer_lock = threading.Lock()
+_orig_encode = PreTrainedTokenizerBase.encode
+
+def _safe_encode(self, *args, **kwargs):
+    with _tokenizer_lock:
+        return _orig_encode(self, *args, **kwargs)
+
+PreTrainedTokenizerBase.encode = _safe_encode
+
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.entrypoints.openai.api_server import run_server
 from vllm.entrypoints.openai.cli_args import make_arg_parser
@@ -49,6 +63,8 @@ if __name__ == "__main__":
         "--enforce-eager",
         "--enable-auto-tool-choice",
         "--tool-call-parser", "hermes",
+        "--disable-frontend-multiprocessing",
+        "--skip-mm-profiling",
     ])
 
     asyncio.run(run_server(args))