Replace TORCH_INTERNAL_ASSERT with TORCH_CHECK in set_history (pytorch#155453)

Fixes pytorch#154357

## Test Result

```bash
>>> import torch
>>>
>>> x = torch.tensor(1, device=torch.device('cpu'))
>>> y = torch.tensor([1.0, 2.0, 3.0], requires_grad=True)
>>> z0 = (x.abs() * y).prod(dtype=torch.int16)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
RuntimeError: Autograd not support dtype: Short
```

Pull Request resolved: pytorch#155453
Approved by: https://github.com/albanD, https://github.com/soulitzer

Author: zeshengzong

test/inductor/test_torchinductor.py

@@ -3872,9 +3872,7 @@ def forward(self, x):
         with self.assertRaisesRegex(RuntimeError, msg):
             with torch.no_grad():
                 torch.compile(fn)(t)
-        # TODO: Autograd internal assertion
-        msg = r".*isDifferentiableType\(variable.scalar_type\(\)\) INTERNAL ASSERT FAILED.*"
-        with self.assertRaisesRegex(RuntimeError, msg):
+        with self.assertRaisesRegex(RuntimeError, "Autograd not support dtype:.*"):
             torch.compile(fn)(t)
 
     @unittest.skipIf(

torch/csrc/autograd/functions/utils.h

@@ -71,7 +71,10 @@ inline void set_history(
     // If the codegen triggers this, you most likely want to add your newly
     // added function to the DONT_REQUIRE_DERIVATIVE list in
     // tools/autograd/gen_variable_type.py
-    TORCH_INTERNAL_ASSERT(isDifferentiableType(variable.scalar_type()));
+    TORCH_CHECK(
+        isDifferentiableType(variable.scalar_type()),
+        "Autograd not support dtype: ",
+        variable.scalar_type());
     auto output_nr = grad_fn->add_input_metadata(variable);
     impl::set_gradient_edge(variable, {grad_fn, output_nr});
   } else {