Fix the problem with "vllm + chatglm3" (lm-sys#2845)

yaofeng · Dec 30, 2023 · 7ef8e43 · 7ef8e43
1 parent 1368f3f
commit 7ef8e43
Showing 1 changed file with 3 additions and 1 deletion.
diff --git a/fastchat/serve/vllm_worker.py b/fastchat/serve/vllm_worker.py
@@ -88,7 +88,9 @@ async def generate_stream(self, params):
 
         for tid in stop_token_ids:
             if tid is not None:
-                stop.add(self.tokenizer.decode(tid))
+                s = self.tokenizer.decode(tid)
+                if s != "":
+                    stop.add(s)
 
         # make sampling params in vllm
         top_p = max(top_p, 1e-5)