Henry Mao

Also, a quick note: @jueseph Replicating your notebook, I got MSE error as such:  https://pytorch.org/docs/stable/generated/torch.nn.MultiheadAttention.html But pytorch's MHA requires tensor to be of [sequence, batch, features],...

@lucidrains Have you tested your FastAttention implementation against JAX implementation (perhaps a unit test e.g. same input tensor), like the one I've written here for my incomplete implementation: https://github.com/calclavia/Performer-Pytorch/blob/main/test.py#L11

@jueseph Tried running your notebook test from a different implementation https://github.com/r0mainK/outperformer/blob/main/src/performer.py   Similar result. Which implies that this could be correct behavior. The...

I'm getting the same issue.

Would be great if interceptors could be async

> I was able to get a WASM module compiled from Rust with wasm-pack working both locally and on Vercel, except in a non-edge API route on Vercel. > >...

Seems related to this issue: https://stackoverflow.com/questions/71254427/firebase-functions-dotenv-variable-undefined