fix code style

Author: eaidova

optimum/exporters/openvino/model_configs.py

@@ -112,9 +112,9 @@ def init_model_configs():
         "transformers",
         "Qwen2VLForConditionalGeneration",
     )
-    TasksManager._TRANSFORMERS_TASKS_TO_MODEL_LOADERS["image-text-to-text"] = (
-        TasksManager._TRANSFORMERS_TASKS_TO_MODEL_LOADERS["text-generation"]
-    )
+    TasksManager._TRANSFORMERS_TASKS_TO_MODEL_LOADERS[
+        "image-text-to-text"
+    ] = TasksManager._TRANSFORMERS_TASKS_TO_MODEL_LOADERS["text-generation"]
 
     supported_model_types = [
         "_SUPPORTED_MODEL_TYPE",

optimum/exporters/openvino/model_patcher.py

@@ -423,9 +423,9 @@ def _llama_gemma_update_causal_mask_legacy(self, attention_mask, input_tensor, c
                 offset = 0
             mask_shape = attention_mask.shape
             mask_slice = (attention_mask.eq(0.0)).to(dtype=dtype) * min_dtype
-            causal_mask[: mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]] = (
-                mask_slice
-            )
+            causal_mask[
+                : mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]
+            ] = mask_slice
 
     if (
         self.config._attn_implementation == "sdpa"
@@ -2060,9 +2060,9 @@ def _dbrx_update_causal_mask_legacy(
                 offset = 0
             mask_shape = attention_mask.shape
             mask_slice = (attention_mask.eq(0.0)).to(dtype=dtype) * min_dtype
-            causal_mask[: mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]] = (
-                mask_slice
-            )
+            causal_mask[
+                : mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]
+            ] = mask_slice
 
     if (
         self.config._attn_implementation == "sdpa"
@@ -3386,10 +3386,9 @@ class Qwen2VLLanguageModelPatcher(DecoderModelPatcher):
     def __init__(
         self,
         config: OnnxConfig,
-        model: PreTrainedModel | TFPreTrainedModel,
-        model_kwargs: Dict[str, Any] | None = None,
+        model: Union[PreTrainedModel, TFPreTrainedModel],
+        model_kwargs: Dict[str, Any] = None,
     ):
-
         model.__orig_forward = model.forward
 
         def forward_wrap(
@@ -3426,8 +3425,8 @@ class Qwen2VLVisionEmbMergerPatcher(ModelPatcher):
     def __init__(
         self,
         config: OnnxConfig,
-        model: PreTrainedModel | TFPreTrainedModel,
-        model_kwargs: Dict[str, Any] | None = None,
+        model: Union[PreTrainedModel, TFPreTrainedModel],
+        model_kwargs: Dict[str, Any] = None,
     ):
         model.__orig_forward = model.forward
 

optimum/exporters/openvino/utils.py

@@ -216,7 +216,15 @@ def get_submodels(model):
     return custom_export, fn_get_submodels
 
 
-MULTI_MODAL_TEXT_GENERATION_MODELS = ["llava", "llava-next", "llava-qwen2", "internvl-chat", "minicpmv", "phi3-v", "qwen2-vl"]
+MULTI_MODAL_TEXT_GENERATION_MODELS = [
+    "llava",
+    "llava-next",
+    "llava-qwen2",
+    "internvl-chat",
+    "minicpmv",
+    "phi3-v",
+    "qwen2-vl",
+]
 
 
 def save_config(config, save_dir):

optimum/intel/openvino/modeling_visual_language.py

@@ -2354,7 +2354,6 @@ def get_multimodal_embeddings(
         video_grid_thw=None,
         **kwargs,
     ):
-
         inputs_embeds = torch.from_numpy(self.get_text_embeddings(input_ids))
         if pixel_values is not None and input_ids.shape[1] != 1:
             image_embeds = torch.from_numpy(self.get_vision_embeddings(pixel_values, image_grid_thw))