update

models/cpp_qwen2.py

@@ -126,12 +126,18 @@ class Qwen2Simulator(Simulator):
             stream=True,
             **self.generation_kwargs
         )
+
+
         generated_text = ""
         # TODO: 检测finish reason，如果是length，则shift，并继续生成。
         for out in output:
             stream = copy.deepcopy(out)
             generated_text += stream["choices"][0]["text"]
-            yield generated_text
+            if "all_text" in stream["choices"][0]:
+                yield stream["choices"][0]["all_text"]
+            else:
+                logger.info("all_text not found")
+                yield generated_text
 
 
 bot = Qwen2Simulator()