KeOps for multitask GPs

[somas193]

I recently tried the KeOps kernels in gpytorch for single task GPs and they work really well, especially from the memory perspective. I also tried wrapping a Multitask kernel around these KeOps kernels and they aren't providing the same kind of memory benefits observed in the single task GP. I was wondering if it's possible to write a custom KeOps kernel for the multitask kernel? Is it doable just by looking at the differences in the normal kernel (say RBF) and KeOps kernel (keops RBF) and then applying these difference to the multitask kernel?