[fbsync] Add test to check that classification models are FX-compatible (#3662)

Summary:
* Add test to check that classification models are FX-compatible

* Replace torch.equal with torch.allclose

* remove skipling

Reviewed By: fmassa

Differential Revision: D29264313

fbshipit-source-id: 4e57e255c6ce680fc6deee6a9980a7d189e23597

Co-authored-by: Nicolas Hug <[email protected]>

Author: NicolasHug

test/test_models.py

@@ -7,6 +7,7 @@
 import functools
 import operator
 import torch
+import torch.fx
 import torch.nn as nn
 import torchvision
 from torchvision import models
@@ -140,6 +141,13 @@ def get_export_import_copy(m):
     assert_export_import_module(sm, args)
 
 
+def _check_fx_compatible(model, inputs):
+    model_fx = torch.fx.symbolic_trace(model)
+    out = model(inputs)
+    out_fx = model_fx(inputs)
+    torch.testing.assert_close(out, out_fx)
+
+
 # If 'unwrapper' is provided it will be called with the script model outputs
 # before they are compared to the eager model outputs. This is useful if the
 # model outputs are different between TorchScript / Eager mode
@@ -408,6 +416,7 @@ def test_classification_model(model_name, dev):
     _assert_expected(out.cpu(), model_name, prec=0.1)
     assert out.shape[-1] == 50
     _check_jit_scriptable(model, (x,), unwrapper=script_model_unwrapper.get(model_name, None))
+    _check_fx_compatible(model, x)
 
     if dev == torch.device("cuda"):
         with torch.cuda.amp.autocast():