make input dynamic and enable sdpa

Author: eaidova

optimum/exporters/openvino/__main__.py

@@ -274,7 +274,11 @@ def main_export(
                 f"Asked to export a {model_type} model for the task {task}{autodetected_message}, but the Optimum OpenVINO exporter only supports the tasks {', '.join(model_tasks.keys())} for {model_type}. Please use a supported task. Please open an issue at https://github.com/huggingface/optimum/issues if you would like the task {task} to be supported in the ONNX export for {model_type}."
             )
 
-        if is_transformers_version(">=", "4.36") and model_type in SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED:
+        if (
+            is_transformers_version(">=", "4.36")
+            and is_transformers_version("<=", "4.45.0")
+            and model_type in SDPA_ARCHS_ONNX_EXPORT_NOT_SUPPORTED
+        ):
             loading_kwargs["attn_implementation"] = "eager"
 
         # some models force flash_attn attention by default that does not support load model on cpu

optimum/exporters/openvino/convert.py

@@ -28,14 +28,14 @@
 from openvino.tools.ovc import convert_model
 from optimum.exporters import TasksManager
 from optimum.exporters.utils import (
-    _get_submodels_and_export_configs as _default_get_submodels_and_export_configs,
-)
-from optimum.exporters.utils import (
-    get_diffusion_models_for_export,
     DECODER_NAME,
     DECODER_WITH_PAST_NAME,
     ENCODER_NAME,
     _get_submodels_for_export_encoder_decoder,
+    get_diffusion_models_for_export,
+)
+from optimum.exporters.utils import (
+    _get_submodels_and_export_configs as _default_get_submodels_and_export_configs,
 )
 from optimum.intel.utils.import_utils import (
     _diffusers_version,

optimum/exporters/openvino/model_configs.py

@@ -2616,6 +2616,13 @@ def patch_model_for_export(
             return StatefulSeq2SeqDecoderPatcher(self, model, model_kwargs)
         return super().patch_model_for_export(model, model_kwargs)
 
+    @property
+    def inputs(self):
+        common_inputs = super().inputs
+        if getattr(self, "stateful", False) and self._behavior == ConfigBehavior.DECODER:
+            common_inputs["decoder_input_ids"] = {0: "batch_size", 1: "seq_length"}
+        return common_inputs
+
 
 @register_in_tasks_manager(
     "t5",
@@ -2630,6 +2637,13 @@ def patch_model_for_export(
             return StatefulSeq2SeqDecoderPatcher(self, model, model_kwargs)
         return super().patch_model_for_export(model, model_kwargs)
 
+    @property
+    def inputs(self):
+        common_inputs = super().inputs
+        if getattr(self, "stateful", False) and self._behavior == ConfigBehavior.DECODER:
+            common_inputs["decoder_input_ids"] = {0: "batch_size", 1: "seq_length"}
+        return common_inputs
+
 
 @register_in_tasks_manager(
     "mt5",

optimum/exporters/openvino/model_patcher.py

@@ -27,8 +27,8 @@
 from optimum.exporters.onnx.model_patcher import (
     DecoderModelPatcher,
     ModelPatcher,
-    override_arguments,
     Seq2SeqModelPatcher,
+    override_arguments,
 )
 from optimum.intel.utils.import_utils import (
     _openvino_version,

optimum/exporters/openvino/stateful.py

@@ -242,7 +242,7 @@ def get_shape_of_ops(model: ov.Model):
 
 def get_consumer_nodes(node):
     consumer_inputs = set().union(*[output.get_target_inputs() for output in node.outputs()])
-    return set(input.get_node() for input in consumer_inputs)
+    return {input.get_node() for input in consumer_inputs}
 
 
 def find_output_nodes_of_dependent_subgraph(model: ov.Model, sources: list):