Fx support for multiple model architectures (#17393)

* Support for Bart and LayoutLM, and partial support for XLNet

* Support for mbart

* A lot of new models supported

* Support for other models

* LayoutLM fix

* Use strings instead of classes

tests/models/trocr/test_modeling_trocr.py

@@ -161,6 +161,7 @@ class TrOCRStandaloneDecoderModelTester:
 class TrOCRStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
     all_model_classes = (TrOCRDecoder, TrOCRForCausalLM) if is_torch_available() else ()
     all_generative_model_classes = (TrOCRForCausalLM,) if is_torch_available() else ()
+    fx_compatible = True
     test_pruning = False
 
     def setUp(self):