add missed imports and disabling quantization

optimum/intel/openvino/modeling_seq2seq.py

@@ -385,7 +385,7 @@ def forward(
         self._compile()
 
         # Model inputs
-        inputs = {self.main_input_name: input_ids}
+        inputs = {self.main_input_name: input_ids if input_ids is not None else kwargs.get(self.main_input_name)}
 
         # Add the attention_mask inputs when needed
         if "attention_mask" in self.input_names:

optimum/intel/openvino/quantization.py

@@ -439,6 +439,9 @@ def _set_task(self):
         if self.task == "text2text-generation":
             raise ValueError("Seq2Seq models are currently not supported for post-training static quantization.")
 
+        if self.task == "image-to-text":
+            raise ValueError("Image2Text models are currently not supported for post-training static quantization.")
+
     def get_calibration_dataset(
         self,
         dataset_name: str,

optimum/intel/openvino/utils.py

@@ -89,6 +89,7 @@
     "audio-classification": "OVModelForAudioClassification",
     "stable-diffusion": "OVStableDiffusionPipeline",
     "stable-diffusion-xl": "OVStableDiffusionXLPipeline",
+    "pix2struct": "OVModelForPix2Struct",
 }
 
 

optimum/intel/utils/constant.py

@@ -22,6 +22,7 @@
     "seq2seq-lm": "text2text-generation",
     "summarization": "text2text-generation",
     "translation": "text2text-generation",
+    "visual-question-answering": "image-to-text",
 }
 
 _TASK_LEGACY = {

optimum/intel/utils/dummy_openvino_objects.py

@@ -103,6 +103,17 @@ def from_pretrained(cls, *args, **kwargs):
         requires_backends(cls, ["openvino"])
 
 
+class OVModelForPix2Struct(metaclass=DummyObject):
+    _backends = ["openvino"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["openvino"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["openvino"])
+
+
 class OVModelForQuestionAnswering(metaclass=DummyObject):
     _backends = ["openvino"]