I have been following this tutorial to implement Customer Op in C++: Extending TorchScript with Custom C++ Operators — PyTorch Tutorials 1.12.0+cu102 documentation
What i wanted to know is that what should the schema look like for registering a custom Op that can take variable no. of arguments ?
Earlier (upto Pytorch1.9) the following types
myCustomOp (Tensor) -> Tensor[]
or
myCustomOp (std::string, Tensor) -> Tensor[]
or
myCustomOp (Tensor, Tensor) -> Tensor[]
and so on could be registered by single schema
myCustomOp(...) -> Tensor
but pytorch 1.10 onwards this is not possible.
Please let me know how should I update my code so I can accept arbitrary no. of arguments like before