- max_new_tokens=1024

+ max_new_tokens=10240
and test

handler.py

@@ -28,7 +28,7 @@ class EndpointHandler:
             bos_token_id=1,
             do_sample=True,
             temperature=0.7,
-            max_new_tokens=1024
+            max_new_tokens=10240
         )
 
     def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
@@ -87,9 +87,15 @@ class EndpointHandler:
                 return_dict_in_generate=True,
                 output_scores=True
             )
-
-        return self.tokenizer.decode(output.sequences[0], skip_special_tokens=True)
-
+            # 🧠 打印 debug 信息
+            generated_ids = output.sequences[0]
+            print("📍生成的 token ids：", generated_ids.tolist())
+            raw_text = self.tokenizer.decode(generated_ids, skip_special_tokens=False)
+            clean_text = self.tokenizer.decode(generated_ids, skip_special_tokens=True)
+            print("🧾 带特殊符号的输出：", raw_text)
+            print("✅ 去掉特殊符号的输出：", clean_text)
+
+            return clean_text  # 返回干净版本
     def postprocess(self, output: str) -> Dict[str, str]:
         return {"output": output.strip()}