Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Probable mistake in Eq. 19 in the arxiv paper "Eagle and Finch" #254

Open
oliverYoung2001 opened this issue Sep 5, 2024 · 3 comments
Open

Comments

@oliverYoung2001
Copy link

oliverYoung2001 commented Sep 5, 2024

Screenshot 2024-09-05 at 11 21 19

The up and down subscripts in equation 19 should be $j=i+1$ and $t-2$ according to the code in "wkv6_cuda.cu".

Screenshot 2024-09-05 at 11 48 04
@BlinkDL
Copy link
Owner

BlinkDL commented Sep 5, 2024

it's u @ t, 1 @ t-1, w @ t-2, w^2 @ t-3, ...

@oliverYoung2001
Copy link
Author

Sorry, I provide the error picture of paper. I mean maybe there are mistakes in "Finch Time Mixing" in the below picture.
Screenshot 2024-09-08 at 14 08 55
Screenshot 2024-09-08 at 14 15 44
Just like codes in "wkv6_cuda.cu", variable $t$ is enumerated in forward order, i.e. traverse all tokens in a forward direction
. Thus every $x_{t}=k_{t}^T @ v_{t}$ will multiply with the $diag(w_{t'})$ after it rather than before it. Thus, maybe the up and down subscripts in equation 19 should be $j=i+1$ and $t-2$ ?

@oliverYoung2001
Copy link
Author

Looking forward to any feedbacks, :)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants