linkedin / linkedin/Liger-Kernel

Multi-token attention

Open
#857 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
6.6k
Forks
603
Avg merge
1d 20h
Merged PRs (30d)
47

Description

### 🚀 The feature, motivation and pitch

Hi! Thank you for adding support for MTA (https://github.com/linkedin/Liger-Kernel/pull/689) ! Do I understand it correctly, that this implementation only covers post-sm key-query convolution? There is also pre-sm Q-K convolution, head convolution, and gated group norm (last one should probably not be part of the kernel). We have released reference code here: https://github.com/facebookresearch/RAM/blob/main/projects/mta/mta_transformer.py#L337

### Alternatives

_No response_

### Additional context

_No response_

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Read the existing MTA support from PR #689 and compare it with the reference implementation in projects/mta/mta_transformer.py around line 337. Clarify which convolution variants are in scope and whether gated group norm should be excluded; done means the implementation scope is agreed and the selected MTA components are supported.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.