Add inputs embeds in generation (#30269)

* Add inputs embeds in generation

* always scale embeds

* fix-copies

* fix failing test

* fix copies once more

* remove embeds for models with scaling

* second try to revert

* codestyle

src/transformers/models/codegen/modeling_codegen.py

@@ -594,7 +594,7 @@ class CodeGenForCausalLM(CodeGenPreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.lm_head = new_embeddings
 
-    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
+    def prepare_inputs_for_generation(self, input_ids, inputs_embeds=None, past_key_values=None, **kwargs):
         token_type_ids = kwargs.get("token_type_ids", None)
         # Omit tokens covered by past_key_values
         if past_key_values:
@@ -621,14 +621,22 @@ class CodeGenForCausalLM(CodeGenPreTrainedModel):
             if past_key_values:
                 position_ids = position_ids[:, -input_ids.shape[1] :]
 
-        return {
-            "input_ids": input_ids,
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids.contiguous()}
+
+        model_inputs.update(
+            {
                 "past_key_values": past_key_values,
                 "use_cache": kwargs.get("use_cache"),
                 "position_ids": position_ids,
                 "attention_mask": attention_mask,
                 "token_type_ids": token_type_ids,
             }
+        )
+        return model_inputs
 
     @add_start_docstrings_to_model_forward(CODEGEN_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(

src/transformers/models/falcon/modeling_falcon.py

@@ -1212,6 +1212,7 @@ class FalconForCausalLM(FalconPreTrainedModel):
         past_key_values: Optional[torch.Tensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
         **kwargs,
     ) -> dict:
         if past_key_values is not None:
@@ -1234,13 +1235,20 @@ class FalconForCausalLM(FalconPreTrainedModel):
             if past_key_values:
                 position_ids = position_ids[:, -input_ids.shape[1] :]
 
-        return {
-            "input_ids": input_ids,
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
                 "position_ids": position_ids,
                 "past_key_values": past_key_values,
                 "use_cache": kwargs.get("use_cache"),
                 "attention_mask": attention_mask,
             }
+        )
+        return model_inputs
 
     @add_start_docstrings_to_model_forward(FALCON_INPUTS_DOCSTRING)
     @add_code_sample_docstrings(

src/transformers/models/gpt2/modeling_gpt2.py

@@ -1430,7 +1430,7 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
     def set_output_embeddings(self, new_embeddings):
         self.lm_head = new_embeddings
 
-    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, **kwargs):
+    def prepare_inputs_for_generation(self, input_ids, inputs_embeds=None, past_key_values=None, **kwargs):
         token_type_ids = kwargs.get("token_type_ids", None)
         # Omit tokens covered by past_key_values
         if past_key_values:
@@ -1459,14 +1459,22 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
         else:
             position_ids = None
 
-        return {
-            "input_ids": input_ids,
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids.contiguous()}
+
+        model_inputs.update(
+            {
                 "past_key_values": past_key_values,
                 "use_cache": kwargs.get("use_cache"),
                 "position_ids": position_ids,
                 "attention_mask": attention_mask,
                 "token_type_ids": token_type_ids,
             }
+        )
+        return model_inputs
 
     @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
     @replace_return_docstrings(output_type=GPT2DoubleHeadsModelOutput, config_class=_CONFIG_FOR_DOC)

src/transformers/models/pegasus/modeling_pegasus.py

@@ -1658,7 +1658,7 @@ class PegasusForCausalLM(PegasusPreTrainedModel):
         )
 
     def prepare_inputs_for_generation(
-        self, input_ids, past_key_values=None, attention_mask=None, use_cache=None, **kwargs
+        self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, use_cache=None, **kwargs
     ):
         # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
         if attention_mask is None:
@@ -1676,12 +1676,19 @@ class PegasusForCausalLM(PegasusPreTrainedModel):
 
             input_ids = input_ids[:, remove_prefix_length:]
         # first step, decoder_cached_states are empty
-        return {
-            "input_ids": input_ids,  # encoder_outputs is defined. input_ids not needed
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids.contiguous()}
+
+        model_inputs.update(
+            {
                 "attention_mask": attention_mask,
                 "past_key_values": past_key_values,
                 "use_cache": use_cache,
             }
+        )
+        return model_inputs
 
     @staticmethod
     def _reorder_cache(past_key_values, beam_idx):