Any way to bypass constraint of input and output features to be a multiple of block size

[shreyanshs]

Hi,

I wanted to know if there is a way to bypass the constraint of the number of input and output features of the Block Sparse Layer being a multiple of a value of the block size. Like is there a generic implementation possible? Something like https://github.com/rain-neuromorphics/SparseLinear which can have any number of input and output features.

I would love to know if there is a way to make it possible?

Thanks