Lightning-AI / Lightning-AI/litgpt

Add support for memory-efficient and faster optimizers

Open
#1,364 1 comment 3 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.7k
Forks
1.5k
Avg merge
15h 37m
Merged PRs (30d)
1

Description

Maybe GaLore (#1192) should be changed from `GaloreArgs` to `OptimizerArgs` after all. Then we can also more easily consider other variants such as BAdam (BAdam: A Memory Efficient Full Parameter Training Method for Large Language Models, https://arxiv.org/abs/2404.02827).

The experiments from [here](https://github.com/Ledzy/BAdam) look very compelling. And it only adds 1 hyperparameter:

Screenshot 2024-04-27 at 8 36 56 AM

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

Start by reviewing the existing GaLore work in issue #1192 and the linked BAdam paper and experiments. Compare the proposed OptimizerArgs approach with the current GaloreArgs design, then define what support for BAdam and other memory-efficient optimizer variants would need to cover. Done means the requested optimizer support and its configuration path are clearly specified and validated.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
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.