fix llava legacy procesing selection

Author: eaidova

optimum/intel/openvino/modeling_visual_language.py

@@ -748,13 +748,12 @@ def merge_vision_text_embeddings(
         image_features = torch.from_numpy(vision_embeds) if isinstance(vision_embeds, np.ndarray) else vision_embeds
         inputs_embeds = torch.from_numpy(inputs_embeds) if isinstance(inputs_embeds, np.ndarray) else inputs_embeds
         if legacy_processing is None:
-            legacy_processing = (
-                not hasattr(self.config, "image_seq_length")
-                or ((input_ids == self.config.image_token_index).sum(1).max() < self.config.image_seq_length)
-                or (input_ids.shape[-1] == 1)
+            legacy_processing = not (hasattr(self.config, "image_seq_length") and (input_ids.shape[-1] == 1)) or (
+                (input_ids == self.config.image_token_index).sum(1).max() < self.config.image_seq_length
             )
 
         if legacy_processing:
+            logger.warn("LEGACY")
             pad_token_id = self.config.pad_token_id if self.config.pad_token_id is not None else -1
 
             num_images, num_image_patches, embed_dim = image_features.shape
@@ -832,11 +831,15 @@ def merge_vision_text_embeddings(
     def get_multimodal_embeddings(
         self, input_ids, pixel_values=None, attention_mask=None, position_ids=None, past_key_values=None, **kwargs
     ):
-        legacy_processing = (
-            not hasattr(self.config, "image_seq_length")
-            or ((input_ids == self.config.image_token_index).sum(1).max() < self.config.image_seq_length)
-            or (input_ids.shape[-1] == 1 and pixel_values is not None)
-        )
+        legacy_processing = getattr(self, "_legacy_processing", not hasattr(self.config, "image_seq_length"))
+        inputs_embeds = self.get_text_embeddings(input_ids, **kwargs)
+
+        if pixel_values is not None and not legacy_processing and past_key_values is None:
+            legacy_processing = (input_ids == self.config.image_token_index).sum(
+                1
+            ).max() < self.config.image_seq_length
+            self._legacy_processing = legacy_processing
+
         inputs_embeds, attention_mask, position_ids = super().get_multimodal_embeddings(
             input_ids, pixel_values, attention_mask, position_ids, legacy_processing=legacy_processing, **kwargs
         )
@@ -847,6 +850,7 @@ def get_multimodal_embeddings(
         return inputs_embeds, attention_mask, position_ids
 
     def _filter_unattended_tokens(self, input_ids, attention_mask, past_key_values):
+        logger.warn("LEGACY")
         if not self.language_model.stateful:
             first_layer_past_key_value = torch.from_numpy(past_key_values[0][0][:, :, :, 0])
         else:
@@ -954,12 +958,14 @@ def get_multimodal_embeddings(
         from transformers.models.llava_next.modeling_llava_next import image_size_to_num_patches
 
         inputs_embeds = self.get_text_embeddings(input_ids, **kwargs)
+        legacy_processing = getattr(self, "_legacy_processing", not hasattr(self.config, "image_seq_length"))
+
+        if pixel_values is not None and not legacy_processing and past_key_values is None:
+            legacy_processing = (input_ids == self.config.image_token_index).sum(
+                1
+            ).max() < self.config.image_seq_length
+            self._legacy_processing = legacy_processing
 
-        legacy_processing = (
-            not hasattr(self.config, "image_seq_length")
-            or ((input_ids == self.config.image_token_index).sum(1).max() < self.config.image_seq_length)
-            or (input_ids.shape[-1] == 1 and pixel_values is not None)
-        )
         if pixel_values is not None and pixel_values.size(0) > 0:
             # ! infer image_num_patches from image_sizes
             image_num_patches = [