Pass encoder outputs into GenerationMixin (#10599)

* Pass encoder_outputs into generate()

* Remove an if-statement

* Reformat

* Minimize changes to generate()

* Comment on input_ids

src/transformers/generation_utils.py

@@ -376,7 +376,14 @@ class GenerationMixin:
         """
         return logits
 
-    def _prepare_input_ids_for_generation(self, bos_token_id: int) -> torch.LongTensor:
+    def _prepare_input_ids_for_generation(
+        self, bos_token_id: Optional[int], encoder_outputs: Optional[ModelOutput]
+    ) -> torch.LongTensor:
+        if self.config.is_encoder_decoder and encoder_outputs is not None:
+            # make dummy input_ids with value -100, as a sanity check ensuring that they won't be used for encoding
+            shape = encoder_outputs.last_hidden_state.size()[:-1]
+            return torch.ones(shape, dtype=torch.long, device=self.device) * -100
+
         if bos_token_id is None:
             raise ValueError("`bos_token_id` has to be defined when no `input_ids` are provided.")
         return torch.ones((1, 1), dtype=torch.long, device=self.device) * bos_token_id
@@ -395,6 +402,7 @@ class GenerationMixin:
     def _prepare_encoder_decoder_kwargs_for_generation(
         self, input_ids: torch.LongTensor, model_kwargs
     ) -> Dict[str, Any]:
+        if "encoder_outputs" not in model_kwargs:
             # retrieve encoder hidden states
             encoder = self.get_encoder()
             encoder_kwargs = {
@@ -887,7 +895,7 @@ class GenerationMixin:
 
         if input_ids is None:
             # init `input_ids` with bos_token_id
-            input_ids = self._prepare_input_ids_for_generation(bos_token_id)
+            input_ids = self._prepare_input_ids_for_generation(bos_token_id, model_kwargs.get("encoder_outputs"))
 
         if model_kwargs.get("attention_mask", None) is None:
             # init `attention_mask` depending on `pad_token_id`