Question about backward scanning

OpenGVLab / VideoMamba

VideoMamba: State Space Model for Efficient Video Understanding

https://arxiv.org/abs/2403.06977

Apache License 2.0

660 stars 47 forks source link

Question about backward scanning #14

Closed SoroushMehraban closed 3 months ago

SoroushMehraban commented 3 months ago

Hi, Thanks for sharing this amazing work. May I know where backward scan is implemented? Looking at here, it seems that only forward scanning is done while Vim does the backward scan by flipping the residuals and hidden states.

Andy1621 commented 3 months ago

Hi! It is done here.

SoroushMehraban commented 3 months ago

That's much better 🙂 Thanks!