You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
parton69 edited this page May 17, 2020
·
7 revisions
SAI, like Leela-Zero, is a distributed effort to train a strong 19x19 Go engine. The learning procedure is based on alternating two steps:
use the current network to play many games;
use the latest games to train a better network.
The first one is the resource-critical step, so we want to do it in a distributed fashion, on as many computers as possible. Step 2 is done on our server and it is not the critical part.
To help the collective effort, you could put your PC up to play games for our server. If you have a powerful computer, you may probably spare part of its resources without feeling differences in your daily use.
Contributed games
The games that need to be played come in two kinds:
self-plays, which are used to train new networks
matches, which are used to evaluate new networks
The self-plays are visible from the SAI page by following the link "self-play games" above the plot.
The matches are visible by following each one of the links "VS" in the "Test Matches" table, just below the plot to the right.