✨[Feature] Support aten::Int.Tensor(Tensor a) -> (int)

[ruoqianguo]

Is your feature request related to a problem? Please describe.

I think "aten::Int" is a common function. Conversion failed because converter aten::Int.Tensor(Tensor a) -> (int) was missing. There is a subgraph about using it.

const auto graph = R"IR(
      graph(%0 : Tensor):
        %1 : int = prim::Constant[value=0]()
        %1.1 : int = prim::Constant[value=1]()
        %1.2 : int = prim::Constant[value=2]()
        %2 : int = aten::size(%0, %1)
        %2.1 : int = aten::size(%0, %1.1)
        %2.2 : int = aten::size(%0, %1.2)
        %3 : Tensor = prim::NumToTensor(%2.1)
        %4 : Tensor = prim::NumToTensor(%2.2)
        %5 : Tensor = aten::mul(%3, %4)
        %6 : int = aten::Int(%5)
        %7 : int[] = prim::ListConstruct(%2, %6)
        %8 : Tensor = aten::view(%0, %7)
        return (%8))IR";

Input could be at::randint(1, 10, {4, 3, 10}, {at::kCUDA})
Describe the solution you'd like

Add a converter.
Describe alternatives you've considered

Additional context

[ErenBalatkan]

I believe this issue occurs when model tries to save static integer information for operations such as reshape. An example would be CSWin Transformer where this information is determined at model creation and stays static always, TorchScript I believe treats such info as IntTensors.

[github-actions]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days