Fix position_ids for infra (#9841)

Author: DrownFish19

llm/predict/predictor.py

@@ -221,16 +221,14 @@ def _preprocess(self, source):
             source = [source] if isinstance(source, str) else source
             source = [self.tokenizer.apply_chat_template(sentence, tokenize=False) for sentence in source]
 
-        return_position_ids = False
         return_attention_mask = False
         if len(source) > 1:
-            return_position_ids = True
             return_attention_mask = True
         tokenized_source = self.tokenizer(
             source,
             max_length=self.config.src_length,
             truncation=True,
-            return_position_ids=True if not isinstance(self.tokenizer, ChatGLMTokenizer) else return_position_ids,
+            return_position_ids=True if not isinstance(self.tokenizer, ChatGLMTokenizer) else False,
             return_attention_mask=return_attention_mask,
             truncation_side="left",
             return_tensors=self.return_tensors,