TalkChess.com

For those wanting to experiment with Leela Chess Zero, but finding the highly optimized, multithreaded c++ code of lc0 too complex, I’ve released Leela Lite, a simple uct algorithm in python. It now has a script that allows it to be used as a uci engine. https://github.com/dkappe/leela_lite Comments...

It is actually one of the path to improvement. Dedicated Alpha / Beta "blunder checker", perhaps running on CPU which usually have spare cycles for LC0. I made one of these for leela in the days of the “blunder prone’ 192x15 networks. You can see the background and results: https://github.com/dkapp...

You may want to use this tool to see how your gpu and cpu compare.

https://github.com/dkappe/leela-ratio

Cpuct to 5 with the random root mod.

It still doesn’t support uci, I presume? Also, any examples of play?

Sorry, the point being that it yields better, more useful results for analysis (and it may play a bit better than vanilla lc0).

I wasn’t too excited about lc0 multipv. It’s very easy for a single move to dominate the visits. I added a random root mod to my bender fork of lc0. After an initial 4000 iterations of search, it explores random root moves 1/16th of the time, then random root moves proportional to Q 3/16ths of the t...

With @dje's no history patch and no pruning, Ender 27 without TB trails only SF on a GTX 1070.
https://github.com/dkappe/leela-chess-w ... ndgame-Net

Thought I'd add two data points: on a GTX 1070, we have the following performance, without TB: Ender 6, Sergei.S.Markoff:Tablebase.Test - 5 sec, 2 threads Success rate: 85.53% (337/394) T947, Sergei.S.Markoff:Tablebase.Test - 5 sec, 2 threads Success rate: 75.38% (297/394) Note that the Leela nets d...