It will take some effort to implement operators but not too much (cutlass's group gemm already support different mnk's), however the performance benefit is marg

It will take some effort to implement operators but not too much (cutlass's group gemm already support different mnk's), however the performance benefit is marginal compared to padding all LoRA ranks to the same rank because all these kernels are not compute bound.