[New Question] Attention with Linear Biases (Medium) #78
Conversation
This file contains hidden or 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
June24-Wu
requested review from
ishaan-arya and
kunal-mansukhani
as code owners
| @@ -0,0 +1,107 @@ | |||
| <p> Implement Attention with Linear Biases (ALiBi) for a given set of matrices. | |||
| Given the query matrix <code>Q</code> of size <code>M×d</code>, key matrix <code>K</code> of size <code>N×d</code>, and value matrix | |||
| <code>V</code> of size <code>N×d</code>, your program should compute the output matrix using the formula: | |||
Contributor
There was a problem hiding this comment.
would be nice to link the paper in the spec
| Given the query matrix <code>Q</code> of size <code>M×d</code>, key matrix <code>K</code> of size <code>N×d</code>, and value matrix | ||
| <code>V</code> of size <code>N×d</code>, your program should compute the output matrix using the formula: | ||
| $$\text{Attention}_{ALiBi}(Q, K, V) = \text{softmax}\Bigl( \frac{QK^T}{\sqrt{d}} + \alpha \cdot \Delta \Bigr)V$$ | ||
| </p> |
Contributor
There was a problem hiding this comment.
52 is taken, do 55
| </p> | ||
|
|
||
| <p> | ||
| where α is a slope controlling the linear bias and <code>Δ = i - j</code> represents the relative position between query <code>i</code> and key <code>j</code>. |
Contributor
There was a problem hiding this comment.
change the dir name from alibi to attn_w_linear_bias
Contributor
Author
There was a problem hiding this comment.
See the new commit to include paper and new dir name
June24-Wu
changed the title
kunal-mansukhani
approved these changes
Sep 16, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
New Question: Attention with Linear Biases
Question is from paper: Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation (https://arxiv.org/pdf/2108.12409)