# Need help debugging torch.compile issue

**URL:** <https://discuss.pytorch.org/t/need-help-debugging-torch-compile-issue/218283>\
**Category:** torch.\_inductor\
**Created:** [March 26, 2025, 9:38am UTC](https://discuss.pytorch.org/t/need-help-debugging-torch-compile-issue/218283 "2025-03-26T09:38:30Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![vivekvpandya](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/vivekvpandya/32/72549_2.png) [@vivekvpandya](https://discuss.pytorch.org/u/vivekvpandya)\
**Post date:** [March 26, 2025, 9:38am UTC](https://discuss.pytorch.org/t/need-help-debugging-torch-compile-issue/218283/1 "2025-03-26T09:38:30Z")

</div>

I have build an experimental torch.compile inductor backend for a custom accelerator.  
Its mostly based on cpu codegen only but I have done some monkey patching which will be removed later.  
It works fine for tests but if I try to use it with cpu inductor backend in same test I see following error:

```auto
    args = (CppCSEVariable(name: tmp0, bounds: VR[-oo, oo], is_vec: False, dtype: torch.float32, dependent_itervars: {x0}),)
kwargs = {}

    def inner(*args, **kwargs):
        bounds = CSEProxy._bound_variable(name, *args, **kwargs)

> value = getattr(parent_handler, name)(*args, **kwargs) # type: ignore[has-type]
E torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
E TypeError: CppOverrides.sigmoid() takes 1 positional argument but 2 were given
E
E Set TORCH_LOGS="+dynamo" and TORCHDYNAMO_VERBOSE=1 for more information
E
E
E You can suppress this exception and fall back to eager by setting:
E import torch._dynamo
E torch._dynamo.config.suppress_errors = True

```

Could someone please guide me on how to debug this further?

Here is my test for reference:

```auto
@pytest.mark.parametrize("dtype", [torch.float, torch.float16])
def test_sigmoid_mean_mul(dtype):
    def sigmoid(x):
        return torch.sigmoid(x)

    def sigmoid_with_mean_reduction(x):
        sig = sigmoid(x)
        mean_sig = torch.mean(sig)
        return mean_sig

    def pointwise_computation(x, mean_sig):
        return x * mean_sig

    def fn(x):
        mean_sig = sigmoid_with_mean_reduction(x)
        result = pointwise_computation(x, mean_sig)
        return result

    x = torch.randn((10, 10000), device="my", dtype=dtype)
    fn_compile = torch.compile(
        fn,
        backend="my-inductor",
        fullgraph=True,
    )
    out_compile = fn_compile(x)
    fn_compile_cpu = torch.compile(
        fn,
        backend="inductor",
        fullgraph=True,
    )
    out = fn_compile_cpu(x.cpu())
    assert torch.allclose(out.cpu(), out_compile.cpu(), rtol=0.01, atol=0.01)

```