[Evidlo]

Aren't there other options for custom PyTorch operators before going so low-level as CUDA C++, like using Jax or CuPy?