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
This file contains 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
* 2018-06-10, ID 396: Switch main training pipeline to use scs-ben's bootstrapped net to try to shortcut fixing [rule50 issues](https://github.com/glinscott/leela-chess/issues/602). Elo dropped for awhile because of this transition, but tests show the rule50 issues are mostly gone.
* 2018-06-05, ID 379: [rule50 plane normalized](https://github.com/glinscott/leela-chess/issues/602). Batch Norm Betas enabled. The main training pipeline is using these changes, but so far recovery from the rule50 norm fix is very slow. In parallel some 3rd parties (scs-ben, logosgg) are bootstrapping a 15x192 and a 20x256 net from existing data. And also the lc0 client is starting from scratch. After we observe progress on these we will decide how to proceed.
* 2018-05-23, ID 330: Generate new networks once every ~40K games, but only do 10K steps (was 50K steps). Revert value_loss_weight back from 0.25 to 1.0.
* 2018-05-13, ID 287: Change value_loss_weight from 1.0 to 0.25 as a temporary workaround for [oversampling](https://github.com/glinscott/leela-chess/wiki/Large-Elo-fluctuations-starting-from-ID253).
There are no files selected for viewing