apply review comments

huggingface · Dec 14, 2023 · 7323be5 · 7323be5
1 parent 2148358
commit 7323be5
Showing 1 changed file with 18 additions and 62 deletions.
diff --git a/optimum/intel/openvino/modeling_seq2seq.py b/optimum/intel/openvino/modeling_seq2seq.py
@@ -627,35 +627,14 @@ def forward(
         past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
         **kwargs,
     ) -> Seq2SeqLMOutput:
-        # Encode if needed : first prediction pass
-        # Encode if needed (training, first prediction pass)
-        if encoder_outputs is None:
-            encoder_outputs = self.encoder(
-                input_ids=flattened_patches,
-                attention_mask=attention_mask,
-            )
-
-        # Decode
-        if past_key_values is None or self.use_cache is False:
-            decoder_outputs = self.decoder(
-                input_ids=decoder_input_ids,
-                decoder_attention_mask=decoder_attention_mask,
-                past_key_values=past_key_values,
-                encoder_hidden_states=encoder_outputs.last_hidden_state,
-                encoder_attention_mask=attention_mask,
-            )
-        else:
-            decoder_outputs = self.decoder_with_past(
-                input_ids=decoder_input_ids[:, -1:],  # Cut decoder_input_ids if past is used
-                decoder_attention_mask=decoder_attention_mask,
-                past_key_values=past_key_values,
-                encoder_hidden_states=encoder_outputs.last_hidden_state,
-                encoder_attention_mask=attention_mask,
-            )
-
-        return Seq2SeqLMOutput(
-            logits=decoder_outputs.logits,
-            past_key_values=decoder_outputs.past_key_values,
+        return super().forward(
+            input_ids=flattened_patches,
+            attention_mask=attention_mask,
+            decoder_input_ids=decoder_input_ids,
+            decoder_attention_mask=decoder_attention_mask,
+            encoder_outputs=encoder_outputs,
+            past_key_values=past_key_values,
+            **kwargs,
         )
 
     def _reshape(self, model: openvino.runtime.Model, batch_size: int, sequence_length: int, is_decoder=True):
@@ -686,7 +665,6 @@ class OVModelForSpeechSeq2Seq(OVModelForSeq2SeqLM):
     def prepare_inputs_for_generation(
         self,
         input_ids,
-        input_features: Optional[torch.FloatTensor] = None,
         attention_mask: Optional[torch.LongTensor] = None,
         decoder_attention_mask: Optional[torch.BoolTensor] = None,
         past_key_values=None,
@@ -701,7 +679,6 @@ def prepare_inputs_for_generation(
             decoder_attention_mask = torch.ones_like(input_ids).to(input_ids.device)
 
         return {
-            "input_features": input_features,
             "decoder_input_ids": input_ids,
             "past_key_values": past_key_values,
             "encoder_outputs": encoder_outputs,
@@ -731,35 +708,14 @@ def forward(
         past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
         **kwargs,
     ) -> Seq2SeqLMOutput:
-        # Encode if needed : first prediction pass
-        # Encode if needed (training, first prediction pass)
-        if encoder_outputs is None:
-            encoder_outputs = self.encoder(
-                input_ids=input_features,
-                attention_mask=attention_mask,
-            )
-
-        # Decode
-        if past_key_values is None or self.use_cache is False:
-            decoder_outputs = self.decoder(
-                input_ids=decoder_input_ids,
-                decoder_attention_mask=decoder_attention_mask,
-                past_key_values=past_key_values,
-                encoder_hidden_states=encoder_outputs.last_hidden_state,
-                encoder_attention_mask=attention_mask,
-            )
-        else:
-            decoder_outputs = self.decoder_with_past(
-                input_ids=decoder_input_ids[:, -1:],  # Cut decoder_input_ids if past is used
-                decoder_attention_mask=decoder_attention_mask,
-                past_key_values=past_key_values,
-                encoder_hidden_states=encoder_outputs.last_hidden_state,
-                encoder_attention_mask=attention_mask,
-            )
-
-        return Seq2SeqLMOutput(
-            logits=decoder_outputs.logits,
-            past_key_values=decoder_outputs.past_key_values,
+        return super().forward(
+            input_ids=input_features,
+            attention_mask=attention_mask,
+            decoder_input_ids=decoder_input_ids,
+            decoder_attention_mask=decoder_attention_mask,
+            encoder_outputs=encoder_outputs,
+            past_key_values=past_key_values,
+            **kwargs,
         )
 
     @classmethod
@@ -770,7 +726,7 @@ def _from_pretrained(
         **kwargs,
     ):
         if "WhisperForConditionalGeneration" in config.architectures:
-            return OVModelForWhisper._from_pretrained(model_id, config, **kwargs)
+            return _OVModelForWhisper._from_pretrained(model_id, config, **kwargs)
         else:
             return super()._from_pretrained(model_id, config, **kwargs)
 
@@ -790,7 +746,7 @@ def mro(cls):
         )
 
 
-class OVModelForWhisper(
+class _OVModelForWhisper(
     OVModelForSpeechSeq2Seq, WhisperForConditionalGeneration, metaclass=MetaClassRemoveParentsAndReorder
 ):
     """