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

Inconsistent initialization of RoPE embedding across component builders #2283

Open
Ankur-singh opened this issue Jan 20, 2025 · 0 comments
Open
Labels
best practice Things we should be doing but aren't better engineering Tasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs

Comments

@Ankur-singh
Copy link
Contributor

Ankur-singh commented Jan 20, 2025

Context

The Llama 3.1 self-attention builder takes RoPE embeddings as an argument, allowing us to build RoPE a single time across all layers. However, the corresponding components for Llama2 and Llama3 do not do this -- they instead construct RoPE for every single layer.

Whether we should use a single global RoPE vs one RoPE per layer? We should standardize this.

Originally posted by @ebsmothers in #2282 (comment)

@felipemello1 felipemello1 added best practice Things we should be doing but aren't better engineering Tasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs labels Jan 22, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
best practice Things we should be doing but aren't better engineering Tasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs
Projects
None yet
Development

No branches or pull requests

2 participants