Implement a "crossmamba" block for different length sequences

[YihangChen9]

I am trying to Implement a "crossmamba" block for different length sequences which may apply in translation or other encoder-decoder task. I had seen several implementations for 2 same length sequences.

I wonder is it pratical to implement such a block? Thanks a lot!

[altaic]

There's some discussion in #229 and #263.

[YihangChen9]

I have seen these discussion, but there is no solution for different length sequence, since it will occur error when during selective scan because of dimensions.

[YihangChen9]

The length dimension of A,B,C should be same, is there any trick to implementaiton,or not yet