chirtee

Do the eig, qr, and svd need to support backpropagation? We notice that the reverse of the complex eig decomposition in PyTorch is ill-conditioned.

Hello, I'm currently contributing to the project and attempting to set up my local development environment. I've encountered some issues while running the tests using `pytest`. Below are the details...

Thank you very much for your patience! Following your previous suggestions, I have installed the `pylatexenc` package and downgraded `torch` to version 2.1. However, I am still encountering several errors....

Any progress on it? I'm looking forward to using NeedleBench to evaluate long-context capabilities better.