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.

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

improve RoPE embeddings #1724

Open
felipemello1 opened this issue Sep 30, 2024 · 0 comments
Open

improve RoPE embeddings #1724

felipemello1 opened this issue Sep 30, 2024 · 0 comments

Comments

@felipemello1
Copy link
Contributor

felipemello1 commented Sep 30, 2024 •
edited
Loading

Opportunities:

  1. Allow users to use rope approximation instead of exact implementation (todo: add link)
  2. Allow flag rope_in_fp32, and if false, then do it in the same dtype as the input (internal only: https://fb.workplace.com/groups/8194229833921360/permalink/8931199086891094/)
  3. Extra: move llama 3.1 rope implementation to the default implementation in modules (https://github.com/pytorch/torchtune/blob/main/torchtune/models/llama3_1/_position_embeddings.py ---> https://github.com/pytorch/torchtune/blob/main/torchtune/modules/position_embeddings.py) @joecummings , can you remind me if you wrote an RFC for this?
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@felipemello1