update

models/cpp_qwen2.py

@@ -47,7 +47,7 @@ class Qwen2Simulator(Simulator):
                 n_ctx=config.MAX_SEQUENCE_LENGTH,  #
                 # n_threads=None, # 默认会根据cpu数来设置 n_threads
                 use_mlock=True,
-                verbose=False,
+                verbose=True,
             )
         else:
             self.hf_tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-0.5B-Instruct")
@@ -132,12 +132,13 @@ class Qwen2Simulator(Simulator):
         # 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"]
-            if "all_text" in stream["choices"][0]:
-                yield stream["choices"][0]["all_text"]
-            else:
-                logger.info("all_text not found")
-                yield generated_text
+            if stream["choices"][0]["finish_reason"] is not None:
+                generated_text += stream["choices"][0]["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()