google-deepmind / google-deepmind/alphageometry

Make AlphaGeometry better, try AG4Masses (https://github.com/tpgh24/ag4masses)

Open
#110 0 comments 4 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
4.9k
Forks
572
PR merge metrics
No merged PRs in 30d

Description

As amazing as AlphaGeometry is, through several months' testing, some limitations of the current version is clear:

- The ability to add auxiliary points is not yet very good compared to humans, as a result it needs a massive amount of computing power to reach IMO level ability (4 GPU V100, 250 CPU). On household hardware it cannot solve many relatively easy problems
- Does not support parallel execution, cannot take advantage of multi-CPU machines
- Limited documentation, difficult for users to test their own problems
- Difficult to setup

I think there is tremendous potential to improve AG and eventually enable it to reach IMO level ability on household hardware. I created a fork repository [AG4Masses](https://github.com/tpgh24/ag4masses) (https://github.com/tpgh24/ag4masses). It currently provides:

- Ability to use multiple CPUs
- Fixed some bugs and cases where AG will crash on some problems
- My experience and tips running AG, including a detailed explanation of its [Problem Definition Language](https://github.com/tpgh24/ag4masses?tab=readme-ov-file#the-problem-definition-language)
- Additional test problems. Logs and analysis of many problems I tested
- Improved code layout, additional tools and detailed setup instructions

I have some ideas on how to further develop AG and bring its power to the masses, to achieve this we need a community. If you are interested, join the [AG4Masses](https://github.com/tpgh24/ag4masses) project!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.