[feature requests] Packing API for linear attention

[John-Ge]

I recently use your awesome flash bidirectional linear attention on my tasks. Since images are of different resolution and native resolution training is critical. I would like to know that if you have plans to develop the varlen API to support packing.
Thank you very much!

[hp-l33]

Thank you for your suggestion, it is indeed a useful feature.

Unfortunately, due to personal matters, I currently have no definite plans to implement this complex project for the time being.

Best