Ignore detokenization error

python/sglang/srt/managers/detokenizer_manager.py

@@ -56,7 +56,7 @@ class DetokenizerManager:
                             int(output_tokens[i][0])
                         )
                         if not isinstance(first_token, str):
-                            first_token = first_token.decode("utf-8")
+                            first_token = first_token.decode("utf-8", errors="ignore")
                         if first_token.startswith("▁"):
                             output_strs[i] = " " + output_strs[i]
 

python/sglang/srt/server.py

@@ -320,7 +320,7 @@ def launch_server(server_args, pipe_finish_writer):
 
     # Load chat template if needed
     if server_args.chat_template is not None:
-        print(server_args.chat_template)
+        print(f"Use chat template: {server_args.chat_template}")
         if not chat_template_exists(server_args.chat_template):
             if not os.path.exists(server_args.chat_template):
                 raise RuntimeError(