[Flax tests/FlaxBert] make from_pretrained test faster (#15561)

tests/test_modeling_flax_bert.py

@@ -141,7 +141,8 @@ class FlaxBertModelTest(FlaxModelTesterMixin, unittest.TestCase):
 
     @slow
     def test_model_from_pretrained(self):
-        for model_class_name in self.all_model_classes:
-            model = model_class_name.from_pretrained("bert-base-cased", from_pt=True)
-            outputs = model(np.ones((1, 1)))
-            self.assertIsNotNone(outputs)
+        # Only check this for base model, not necessary for all model classes.
+        # This will also help speed-up tests.
+        model = FlaxBertModel.from_pretrained("bert-base-cased")
+        outputs = model(np.ones((1, 1)))
+        self.assertIsNotNone(outputs)