diff --git a/test/test_meta.py b/test/test_meta.py
index 29fc07a9eee..fef1d0df506 100644
--- a/test/test_meta.py
+++ b/test/test_meta.py
@@ -17,6 +17,7 @@ from torch.testing._internal.common_utils import (
     skipIfTorchDynamo,
     suppress_warnings,
     TEST_WITH_ASAN,
+    TEST_WITH_TORCHDYNAMO,
     run_tests,
     dtype_abbrs,
     parametrize
@@ -1105,6 +1106,13 @@ class TestMeta(TestCase):
     @suppress_warnings
     @ops(op_db)
     def test_meta_outplace(self, device, dtype, op):
+        skip_op_names = (
+            "fft.ihfft",
+            "fft.ihfft2",
+            "linalg.lu_solve",
+        )
+        if TEST_WITH_TORCHDYNAMO and op.name in skip_op_names:
+            raise unittest.SkipTest("flaky")
         # run the OpInfo sample inputs, cross-referencing them with the
         # meta implementation and check the results are the same.  All
         # the heavy lifting happens in MetaCrossRefFunctionMode
diff --git a/test/test_nestedtensor.py b/test/test_nestedtensor.py
index b9ee51ed41f..f82523a514b 100644
--- a/test/test_nestedtensor.py
+++ b/test/test_nestedtensor.py
@@ -26,6 +26,7 @@ from torch.testing._internal.common_utils import (
     parametrize,
     run_tests,
     skipIfSlowGradcheckEnv,
+    skipIfTorchDynamo,
     subtest,
     TestCase,
 )
@@ -934,6 +935,7 @@ class TestNestedTensorDeviceType(TestCase):
         is_cuda = 'cuda' in str(device)
         self.assertEqual(nt.is_cuda, is_cuda)
 
+    @skipIfTorchDynamo("flaky")
     @dtypes(torch.float, torch.float16, torch.double)
     def test_nested_tensor_indexing(self, device, dtype):
         # edge case: empty nested tensor
diff --git a/test/test_ops.py b/test/test_ops.py
index f69d0a369c5..96b5616aa30 100644
--- a/test/test_ops.py
+++ b/test/test_ops.py
@@ -34,6 +34,7 @@ from torch.testing._internal.common_utils import (
     suppress_warnings,
     noncontiguous_like,
     TEST_WITH_ASAN,
+    TEST_WITH_TORCHDYNAMO,
     TEST_WITH_UBSAN,
     IS_WINDOWS,
     IS_FBCODE,
@@ -651,6 +652,8 @@ class TestCommon(TestCase):
     #   - out= with the correct dtype and device, but the wrong shape
     @ops(ops_and_refs, dtypes=OpDTypes.none)
     def test_out_warning(self, device, op):
+        if TEST_WITH_TORCHDYNAMO and op.name == "_refs.clamp":
+            self.skipTest("flaky")
         # Prefers running in float32 but has a fallback for the first listed supported dtype
         supported_dtypes = op.supported_dtypes(self.device_type)
         if len(supported_dtypes) == 0: