Send attention_mask to device

#9
by mverrilli - opened
Databricks org

attention_mask to be sent to device if provided

Databricks org

I needed this for MPS (not that MPS works well, but that's a different issue). Would this be required for other devices?

Databricks org

@matthayes seems reasonable to me, any concerns? we'd have to do it for the other models

Databricks org

This looks fine to me. I tested this and did not see any problems with it.

matthayes changed pull request status to merged

Sign up or log in to comment