Add visualbert onnx suport

optimum/exporters/onnx/model_configs.py

@@ -118,6 +118,25 @@ def inputs(self) -> Dict[str, Dict[int, str]]:
         }
 
 
+class VisualBertOnnxConfig(TextAndVisionOnnxConfig):
+    DEFAULT_ONNX_OPSET = 11
+
+    @property
+    def inputs(self) -> Dict[str, Dict[int, str]]:
+        return {
+            "input_ids": {0: "batch_size", 1: "sequence_length"},
+            "attention_mask": {0: "batch_size", 1: "sequence_length"},
+            "pixel_values": {0: "batch_size", 1: "num_channels", 2: "height", 3: "width"},
+        }
+
+    @property
+    def outputs(self) -> Dict[str, Dict[int, str]]:
+        return {
+            "last_hidden_state": {0: "batch_size", 1: "sequence_length"},
+            "pooler_output": {0: "batch_size"},
+        }
+
+
 class AlbertOnnxConfig(BertOnnxConfig):
     DEFAULT_ONNX_OPSET = 14  # now uses F.scaled_dot_product_attention by default for torch>=2.1.1.
 

optimum/exporters/tasks.py

@@ -1108,6 +1108,12 @@ class TasksManager:
             "text-to-audio",
             onnx="VitsOnnxConfig",
         ),
+        "visualbert": supported_tasks_mapping(
+            "multiple-choice",
+            "question-answering",
+            "image-to-text",
+            onnx="VisualBertOnnxConfig",
+        ),
         "wavlm": supported_tasks_mapping(
             "feature-extraction",
             "automatic-speech-recognition",

tests/exporters/exporters_utils.py

@@ -197,6 +197,7 @@
             "document-question-answering-with-past",
         ],
     },
+    "visualbert": "hf-internal-testing/tiny-random-VisualBertModel",
 }
 
 
@@ -286,6 +287,7 @@
     "speech-to-text": "codenamewei/speech-to-text",
     "xlm": "xlm-clm-ende-1024",
     "xlm-roberta": "Unbabel/xlm-roberta-comet-small",
+    "visualbert": "unc-nlp/visualbert-uncased",
 }
 
 TENSORFLOW_EXPORT_MODELS = {