[HOWTO failed] Can blocksparse attention support attention_mask?

[SkyAndCloud]

Hi, I run the blocksparse code snippet in HOWTO.md and failed. The error message shows:

AssertionError: This attention does not support attention masks

This is caused by line. It seems that there needs to be improved for HOWTO.md. Futhermore, we may need a flexible sparse attention which supports arbitrary specefic attention mask in many cases. It would be wonderful and powerful. Can blocksparse attention support attention_mask?

[fuvty]

I run into the same problem. Looking forward to any possible solutions.

[nofreewill42]

Still the same problem