update

models/cpp_qwen2.py

@@ -69,7 +69,7 @@ class Qwen2Simulator(Simulator):
             "<|im_start|>",
             "<|endoftext|>",
         ]
-        self.stop_tokens = self.tokenize(self.stop_words)
+        self.stop_tokens = self.tokenize("".joself.stop_words)
         self.generation_kwargs = dict(
             temperature=config.DEFAULT_TEMPERATURE,
             top_p=config.DEFAULT_TOP_P,
@@ -91,6 +91,7 @@ class Qwen2Simulator(Simulator):
             tokens.pop(0)
         while tokens and tokens[-1] in self.stop_tokens:
             tokens.pop()
+        return tokens
 
     def generate(self, history, stream=True):
         if history[-1]['role'] in ["user"]:
@@ -102,8 +103,8 @@ class Qwen2Simulator(Simulator):
         for message in history:
             if "tokens" not in message:  # tokens
                 message["tokens"] = self.tokenize(message["content"])
-            input_ids += self._strip_stoptokens(self.tokenize(f"<|im_start|>{message['role']}\n")) \
-                         + message["tokens"] \
+            input_ids += self.tokenize(f"<|im_start|>{message['role']}\n") \
+                         + self._strip_stoptokens(message["tokens"]) \
                          + self.tokenize("<|im_end|>\n")
         input_ids += start_tokens
         if stream: