Update modeling_codeshell.py

modeling_codeshell.py

@@ -970,7 +970,7 @@ class CodeShellForCausalLM(CodeShellPreTrainedModel):
         generation_config = generation_config or self.generation_config
         input_ids = self.build_chat_input(query, history, tokenizer, generation_config.max_new_tokens)
         stopping_criteria = StoppingCriteriaList(
-            [EndOfFunctionCriteria([len(input_ids[0])], ['|<end>|', '|end|', '<|endoftext|>'], tokenizer)]
+            [EndOfFunctionCriteria([len(input_ids[0])], ['|<end>|', '|end|', '<|endoftext|>', '## human'], tokenizer)]
         )
         
         if stream:
@@ -994,7 +994,7 @@ class CodeShellForCausalLM(CodeShellPreTrainedModel):
         input_ids = input_ids[-max_input_tokens:]  # truncate left
 
         stopping_criteria = StoppingCriteriaList(
-            [EndOfFunctionCriteria([len(input_ids[0])], ['|end|', '|<end>|', '<|endoftext|>'], tokenizer)]
+            [EndOfFunctionCriteria([len(input_ids[0])], ['|<end>|', '|end|', '<|endoftext|>', '## human'], tokenizer)]
         )
 
         streamer = TextIterStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)