fix template.decode (#3497)

Author: Jintao-Huang

swift/llm/infer/infer_engine/utils.py

@@ -86,7 +86,7 @@ def _get_response(self, response: str, is_finished: bool, token_len: int) -> str
     def get_printable_text(self, raw_tokens: List[int], is_finished: bool) -> str:
         raw_tokens = raw_tokens[self.cache_idx:]
         response = self.template.decode(
-            raw_tokens, is_finished, tokenizer_kwargs=self.decode_kwargs, first_token=self.first_token)
+            raw_tokens, is_finished=is_finished, tokenizer_kwargs=self.decode_kwargs, first_token=self.first_token)
         self.first_token = False
         response = self._align_blank_suffix(response)
         return self._get_response(response, is_finished, len(raw_tokens))

swift/llm/template/base.py

@@ -359,8 +359,8 @@ def decode_seq_cls(self, logits: torch.Tensor):
 
     def decode(self,
                generate_ids: List[int],
-               is_finished: bool = True,
                *,
+               is_finished: bool = True,
                tokenizer_kwargs=None,
                first_token=True,
                **kwargs) -> Any: