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.

Sign up for GitHub

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

Jump to bottom

Compressed Multi-head attention as embedding #125

Open

Bigfield77 opened this issue Jul 1, 2024 · 0 comments

Bigfield77 commented Jul 1, 2024

Hello

Looking at llama 3-8b, it has 32 heads of (1,32,7,7) so around 50176 terms.

If this was flattened and compressed somehow to the input required for Pixart Sigma (300 tokens?)

like a latent text embedding space, could this be used to train a model from scratch?
would this be any good?

I guess masking would need to be changed?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment