[fbsync] Use torch.testing.assert_close in test_models.py (#3879)

Summary: Co-authored-by: Philip Meier <github.pmeier@posteo.de>

Reviewed By: vincentqb, cpuhrsch

Differential Revision: D28679965

fbshipit-source-id: 18f1fb6e548afbb1d3b97ec88cc5d3f6d4699582

test/test_models.py

@@ -120,7 +120,7 @@ def check_out(out):
                 # predictions match.
                 expected_file = self._get_expected_file(name)
                 expected = torch.load(expected_file)
-                self.assertEqual(out.argmax(dim=1), expected.argmax(dim=1), prec=prec)
+                torch.testing.assert_close(out.argmax(dim=1), expected.argmax(dim=1), rtol=prec, atol=prec)
                 return False  # Partial validation performed
 
             return True  # Full validation performed
@@ -205,7 +205,8 @@ def compute_mean_std(tensor):
                 # scores.
                 expected_file = self._get_expected_file(name)
                 expected = torch.load(expected_file)
-                self.assertEqual(output[0]["scores"], expected[0]["scores"], prec=prec)
+                torch.testing.assert_close(output[0]["scores"], expected[0]["scores"], rtol=prec, atol=prec,
+                                           check_device=False, check_dtype=False)
 
                 # Note: Fmassa proposed turning off NMS by adapting the threshold
                 # and then using the Hungarian algorithm as in DETR to find the
@@ -301,10 +302,8 @@ def test_memory_efficient_densenet(self):
             model2.eval()
             out2 = model2(x)
 
-            max_diff = (out1 - out2).abs().max()
-
             self.assertTrue(num_params == num_grad)
-            self.assertTrue(max_diff < 1e-5)
+            torch.testing.assert_close(out1, out2, rtol=0.0, atol=1e-5)
 
     def test_resnet_dilation(self):
         # TODO improve tests to also check that each layer has the right dimensionality