Fix unintended updates to submodules

torch/_dynamo/variables/builtin.py

@@ -1991,7 +1991,7 @@ class BuiltinVariable(VariableTracker):
             # If the object implements a __getitem__ method, iter(...) will call obj.__getitem__()
             # with an integer argument starting at 0, until __getitem__ raises IndexError
             ret = variables.UserFunctionVariable(
-                polyfills.builtins.iter_
+                polyfills.builtins.iter_  # type: ignore[arg-type]
             ).call_function(tx, [obj, *args], {})
 
             if args:

torch/_dynamo/variables/iter.py

@@ -590,7 +590,7 @@ class FilterVariable(IteratorVariable):
             else:
                 res = self.fn.call_function(tx, [item], {})
             pred_res = variables.UserFunctionVariable(
-                polyfills.predicate
+                polyfills.predicate  # type: ignore[arg-type]
             ).call_function(tx, [res], {})
             if pred_res.as_python_constant():
                 return item

torch/_dynamo/variables/lists.py

@@ -1498,6 +1498,7 @@ class NamedTupleVariable(TupleVariable):
                     variables.UserDefinedClassVariable(self.tuple_cls),
                 )
             elif isinstance(method, staticmethod):
+                # pyrefly: ignore[bad-argument-type]
                 return UserFunctionVariable(method.__func__)
             elif inspect.isfunction(method):
                 return UserMethodVariable(method, self)

torch/_dynamo/variables/torch.py

@@ -472,7 +472,12 @@ class TorchCtxManagerClassVariable(BaseTorchVariable):
             )
         elif self.value is torch.nn.attention.sdpa_kernel.__wrapped__:  # type: ignore[attr-defined]
             name_to_arg_map = bind_args_cached(
-                self.value, tx, self.source, args, kwargs
+                # pyrefly: ignore[bad-argument-type]
+                self.value,
+                tx,
+                self.source,
+                args,
+                kwargs,
             )
             backends = name_to_arg_map["backends"].as_python_constant()
             set_priority = name_to_arg_map["set_priority"].as_python_constant()
@@ -1429,7 +1434,7 @@ class TorchInGraphFunctionVariable(BaseTorchVariable):
             packed_input_vt = TupleVariable.build(
                 tx, (TupleVariable.build(tx, args), ConstDictVariable.build(tx, kwargs))
             )
-            out_vt = variables.UserFunctionVariable(tree_flatten).call_function(
+            out_vt = variables.UserFunctionVariable(tree_flatten).call_function(  # type: ignore[arg-type]
                 tx, [packed_input_vt], {}
             )
             assert isinstance(out_vt, TupleVariable) and len(out_vt.items) == 2