improve test robustness for Normalize (meta-pytorch#1518)

Summary:
Pull Request resolved: meta-pytorch#1518

see title. This bumps the tolerances. This was flaky locally.

Differential Revision: https://internalfb.com/D41482368

fbshipit-source-id: 75b25a85e3e1b8996c06e36045924d340607c941

Author: sdaulton

test/models/transforms/test_input.py

@@ -232,7 +232,7 @@ def test_normalize(self):
 
                 nlz.eval()
                 X_unnlzd = nlz.untransform(X_nlzd)
-                self.assertAllClose(X, X_unnlzd, atol=1e-4, rtol=1e-4)
+                self.assertAllClose(X, X_unnlzd, atol=1e-3, rtol=1e-3)
                 expected_bounds = torch.cat(
                     [X.min(dim=-2, keepdim=True)[0], X.max(dim=-2, keepdim=True)[0]],
                     dim=-2,