Thank you for releasing the code and for making it well documented! I'm really fascinated by this recent line of work on Atom3D, GVP, and the recent vector gating extension.
I was wondering if vector gating always improves performance over the original GVP design, even for tasks from the ICLR'21 paper?
Thank you for releasing the code and for making it well documented! I'm really fascinated by this recent line of work on Atom3D, GVP, and the recent vector gating extension.
I was wondering if vector gating always improves performance over the original GVP design, even for tasks from the ICLR'21 paper?