update

models/cpp_qwen2.py

@@ -197,10 +197,12 @@ class Qwen2Simulator(Simulator):
             else:
                 logger.info(f'finish_reason {stream["choices"][0]["finish_reason"]} with text: {stream["choices"][0]["text"]}')
 
-        # warmup for next turn
+        # warmup for next turn (下轮解码的加速)
         if suffix_tokens:
             # <|im_end|>\n
-            self.llm.eval([151645, 198] + suffix_tokens)  # 增加 n_tokens，
+            logger.info(f"before warmup: n_tokens = {self.llm.n_tokens}")
+            self.llm.eval([151645, 198] + suffix_tokens)  # 增加 n_tokens
+            logger.info(f"after warmup: n_tokens = {self.llm.n_tokens}")