update

models/cpp_qwen2.py

@@ -129,13 +129,16 @@ if __name__ == "__main__":
     # print(output)
 
     messages = [{"role": "system", "content": "你是一个导游。"}]
+    generated_tokens = None
     print("######## requesting", messages)
-    for generated_text, generated_tokens in bot.generate_query(messages, stream=True):
+    for generated_text, generated_tokens in bot.generate(messages, stream=True):
         print(generated_text, generated_tokens)
 
+
     for i in range(3):
         messages.append({"role": "user" if i % 2 == 0 else "assistant", "content": generated_text, "tokens": generated_tokens})
         print("######## requesting", messages)
-        for generated_text, all_tokens in bot.generate(messages, stream=True):
-            print(generated_text, all_tokens)
+        for generated_text, generated_tokens in bot.generate(messages, stream=True):
+            pass
+            # print(generated_text, all_tokens)