[Generation] Fix bug for manual decoder_input_ids + warning message (#9472)

* up

* improve style

src/transformers/generation_utils.py

@@ -379,12 +379,8 @@ class GenerationMixin:
         return model_kwargs
 
     def _prepare_decoder_input_ids_for_generation(
-        self, input_ids: torch.LongTensor, decoder_start_token_id: int = None, bos_token_id: int = None, **model_kwargs
+        self, input_ids: torch.LongTensor, decoder_start_token_id: int = None, bos_token_id: int = None
     ) -> torch.LongTensor:
-
-        if "decoder_input_ids" in model_kwargs:
-            return model_kwargs["decoder_input_ids"]
-
         decoder_start_token_id = self._get_decoder_start_token_id(decoder_start_token_id, bos_token_id)
         decoder_input_ids = (
             torch.ones((input_ids.shape[0], 1), dtype=input_ids.dtype, device=input_ids.device)
@@ -837,13 +833,23 @@ class GenerationMixin:
             model_kwargs = self._prepare_encoder_decoder_kwargs_for_generation(input_ids, model_kwargs)
 
             # set input_ids as decoder_input_ids
+            if "decoder_input_ids" in model_kwargs:
+                input_ids = model_kwargs.pop("decoder_input_ids")
+            else:
                 input_ids = self._prepare_decoder_input_ids_for_generation(
-                input_ids, decoder_start_token_id=decoder_start_token_id, bos_token_id=bos_token_id, **model_kwargs
+                    input_ids, decoder_start_token_id=decoder_start_token_id, bos_token_id=bos_token_id
                 )
 
             if "encoder_outputs" not in model_kwargs or not isinstance(model_kwargs["encoder_outputs"], ModelOutput):
                 raise ValueError("Make sure that `model_kwargs` include `encoder_outputs` of type `ModelOutput`.")
 
+        if input_ids.shape[-1] >= max_length:
+            input_ids_string = "decoder_input_ids" if self.config.is_encoder_decoder else "input_ids"
+            logger.warning(
+                f"Input length of {input_ids_string} is {input_ids.shape[-1]}, but ``max_length`` is set to {max_length}."
+                "This can lead to unexpected behavior. You should consider increasing ``config.max_length`` or ``max_length``."
+            )
+
         # determine generation mode
         is_greedy_gen_mode = (num_beams == 1) and (num_beam_groups == 1) and do_sample is False
         is_sample_gen_mode = (num_beams == 1) and (num_beam_groups == 1) and do_sample is True