Avoid duplicate input kwargs in `_decode`

modeling_minicpmo.py

@@ -649,6 +649,7 @@ class MiniCPMO(MiniCPMOPreTrainedModel):
         return outputs
 
     def _decode_stream(self, inputs_embeds, tokenizer, **kwargs):
+        kwargs.pop("output_hidden_states", None)
         terminators = [tokenizer.convert_tokens_to_ids(i) for i in self.terminators]
         streamer = TextIteratorStreamer(tokenizer=tokenizer)
         generation_kwargs = {
@@ -777,6 +778,7 @@ class MiniCPMO(MiniCPMOPreTrainedModel):
         tokenizer=None,
         vision_hidden_states=None,
         stream=False,
+        return_dict_in_generate=False,
         **kwargs,
     ):
         assert input_ids is not None
@@ -814,7 +816,10 @@ class MiniCPMO(MiniCPMOPreTrainedModel):
                 outputs = self._decode(model_inputs["inputs_embeds"], tokenizer, attention_mask, **kwargs)
 
                 result = self._decode_text(outputs.sequences, tokenizer)
-
+                
+        if return_dict_in_generate is True:
+            return outputs
+        
         return result, outputs
 
     def chat(