pytorch
diff --git a/‎test/dynamo/test_modules.py
Lines changed: 0 additions & 48 deletions b/‎test/dynamo/test_modules.py
Lines changed: 0 additions & 48 deletions
diff --git a/‎torch/_dynamo/config.py
Lines changed: 0 additions & 4 deletions b/‎torch/_dynamo/config.py
Lines changed: 0 additions & 4 deletions
diff --git a/‎torch/csrc/dynamo/guards.cpp
Lines changed: 1 addition & 7 deletions b/‎torch/csrc/dynamo/guards.cpp
Lines changed: 1 addition & 7 deletions
@@ -3136,54 +3136,6 @@ def fn(x):
         res = opt_fn(x)
         self.assertEqual(ref, res)
 
-    @patch.object(
-        torch._dynamo.config, "skip_tensor_guards_with_matching_dict_tags", False
-    )
-    def test_param_requires_grad(self):
-        def adjust_model(model):
-            to_freeze = model.num_iter % 2 == 0
-            if to_freeze:
-                for param in model.layer2.parameters():
-                    param.requires_grad = False
-            else:
-                for param in model.layer2.parameters():
-                    param.requires_grad = True
-
-        class MyModule(torch.nn.Module):
-            def __init__(self, input_size, hidden_size, output_size):
-                super().__init__()
-
-                self.layer1 = torch.nn.Linear(hidden_size, hidden_size)
-                self.layer2 = torch.nn.Linear(hidden_size, hidden_size)
-
-                self.num_iter = 0
-
-            def forward(self, x):
-                x = self.layer2(x + self.layer1.bias)
-
-                self.num_iter += 1
-                return x
-
-        input_size = 1024
-        hidden_size = 1024
-        output_size = 1
-        num_samples = 2048
-        features = torch.randn(num_samples, input_size)
-
-        model = MyModule(input_size, hidden_size, output_size)
-
-        cnt = torch._dynamo.testing.CompileCounter()
-        opt_model = torch.compile(model, backend=cnt, fullgraph=True)
-
-        for _ in range(3):
-            model.zero_grad(True)
-            adjust_model(model)
-            res = opt_model(features)
-            res.sum().backward()
-
-        # Check that we have recompiled twice, which leads to 3 frames
-        self.assertEqual(cnt.frame_count, 3)
-
 
 if __name__ == "__main__":
     from torch._dynamo.test_case import run_tests
 
@@ -331,10 +331,6 @@ def _get_optimize_ddp_mode():
 # notice and lead to incorrect result.
 skip_no_tensor_aliasing_guards_on_parameters = True
 
-# Considers a tensor immutable if it is one of the values of a dictionary, and
-# the dictionary tag is same across invocation calls.
-skip_tensor_guards_with_matching_dict_tags = True
-
 # If True, raises exception if TorchDynamo is called with a context manager
 raise_on_ctx_manager_usage = True
 
 
@@ -886,11 +886,6 @@ std::string get_exception_message() {
 }
 
 bool is_immutable_object(py::handle example_value) {
-  static py::object config_module = py::module_::import("torch._dynamo.config");
-  bool is_tensor_immutable =
-      config_module.attr("skip_tensor_guards_with_matching_dict_tags")
-          .cast<bool>();
-
   if (PyTupl
887A
e_Check(example_value.ptr())) {
     // Check that each element is immutable
     for (Py_ssize_t i = 0; i < PyTuple_Size(example_value.ptr()); ++i) {
@@ -901,11 +896,10 @@ bool is_immutable_object(py::handle example_value) {
     }
     return true;
   }
-
   return PyLong_Check(example_value.ptr()) ||
       PyFloat_Check(example_value.ptr()) || PyBool_Check(example_value.ptr()) ||
       PyUnicode_Check(example_value.ptr()) ||
-      (is_tensor_immutable && THPVariable_Check(example_value.ptr()));
+      THPVariable_Check(example_value.ptr());
 }
 
 bool is_parameter(py::handle tensor) {
Original file line number	Diff line number	Diff line change
`@@ -886,11 +886,6 @@ std::string get_exception_message() {`
`886`	`886`	`}`
`887`	`887`
`888`	`888`	`bool is_immutable_object(py::handle example_value) {`
`889`		`- static py::object config_module = py::module_::import("torch._dynamo.config");`
`890`		`- bool is_tensor_immutable =`
`891`		`- config_module.attr("skip_tensor_guards_with_matching_dict_tags")`
`892`		`- .cast<bool>();`
`893`		`-`
`894`	`889`	`if (PyTupl 887A e_Check(example_value.ptr())) {`
`895`	`890`	`// Check that each element is immutable`
`896`	`891`	`for (Py_ssize_t i = 0; i < PyTuple_Size(example_value.ptr()); ++i) {`
`@@ -901,11 +896,10 @@ bool is_immutable_object(py::handle example_value) {`
`901`	`896`	`}`
`902`	`897`	`return true;`
`903`	`898`	`}`
`904`		`-`
`905`	`899`	`return PyLong_Check(example_value.ptr()) \|\|`
`906`	`900`	`PyFloat_Check(example_value.ptr()) \|\| PyBool_Check(example_value.ptr()) \|\|`
`907`	`901`	`PyUnicode_Check(example_value.ptr()) \|\|`
`908`		`- (is_tensor_immutable && THPVariable_Check(example_value.ptr()));`
	`902`	`+ THPVariable_Check(example_value.ptr());`
`909`	`903`	`}`
`910`	`904`
`911`	`905`	`bool is_parameter(py::handle tensor) {`