diff --git a/tests/test_modules.py b/tests/test_modules.py
index f08e0e470e..b382106ef0 100755
--- a/tests/test_modules.py
+++ b/tests/test_modules.py
@@ -196,6 +196,7 @@ class NNModuleTests(torchdynamo.testing.TestCase):
     test_iseval2 = make_test(IsEvalCheck())
     test_viamodulecall = make_test(ViaModuleCall())
     test_isnonelayer = make_test(IsNoneLayer())
+    test_intarg = make_test(IntArg())
 
     # not yet implemented
     # test_layerlist = make_test(LayerList())
diff --git a/torchdynamo/symbolic_convert.py b/torchdynamo/symbolic_convert.py
index c375f46e71..4ce860cac4 100644
--- a/torchdynamo/symbolic_convert.py
+++ b/torchdynamo/symbolic_convert.py
@@ -4,6 +4,7 @@
 import functools
 import inspect
 import itertools
+from numbers import Real
 import operator
 import types
 import typing
@@ -23,6 +24,7 @@
 from .guards import GuardSource
 from .variable_tracker import AllowedFunctionOrModuleVariable, PythonModuleVariable
 from .variable_tracker import BaseListVariable
+from .variable_tracker import BasicTypeVariable
 from .variable_tracker import BuiltinVariable
 from .variable_tracker import ConstDictVariable
 from .variable_tracker import ConstantVariable
@@ -184,6 +186,13 @@ def wrap_local(self, name, value):
                 value=value,
                 guards={Guard(name, GuardSource.LOCAL, GuardBuilder.VALUE_MATCH)},
             )
+        elif isinstance(value, Real):
+            self.graphargs.append(LocalArg(name))
+            return BasicTypeVariable(
+                proxy=self.create_graph_input(name),
+                state=TracingSupported.UNKNOWN,
+                guards={Guard(name, GuardSource.LOCAL, GuardBuilder.TYPE_MATCH)},
+            )
         elif type(value) in (tuple, list) and all(
             isinstance(x, torch.Tensor) for x in value
         ):
diff --git a/torchdynamo/variable_tracker.py b/torchdynamo/variable_tracker.py
index 2914886ad2..3f95d4ecc4 100644
--- a/torchdynamo/variable_tracker.py
+++ b/torchdynamo/variable_tracker.py
@@ -97,6 +97,15 @@ def as_proxy(self):
         return self.proxy
 
 
+class BasicTypeVariable(TensorVariable):
+    """
+    Points to a simple type, e.g. int, float, str. So far, we treat this
+    the same as TensorVariable
+    """
+
+    pass
+
+
 class NNModuleVariable(VariableTracker):
     def __init__(self, module_key: str, **kwargs):
         super(NNModuleVariable, self).__init__(**kwargs)