Adding support for Rotary Position Embeddings #675
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Pulling the changes from main branch
|
@dakinggg will this merge cause any problems in merging mpt code into the Hugging Face github codebase? |
vchiley
approved these changes
Oct 31, 2023
dakinggg
reviewed
Nov 1, 2023
dakinggg
approved these changes
Nov 4, 2023
Co-authored-by: Daniel King <[email protected]>
Co-authored-by: Daniel King <[email protected]>
Co-authored-by: Daniel King <[email protected]>
Co-authored-by: Daniel King <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Adding the support for Rotary Positional Embeddings(RoPE).
tl;dr: Advantages of RoPE: This embedding gets applied to the query and key matrices, and hence (unlike ALiBi embeddings), is agnostic to the attention implementation and works out-of-the box for any attention implementation. Recent works have also shown that some variants of RoPE are good at extrapolating beyond training length.
Design doc
Experiments: 125M model, 1B model.