Edit on GitHub

Where to play with Lc0

Getting Started

To set up Leela Chess Zero, follow the following steps:

1. Setup a chess GUI

Leela Chess Zero, like most other chess programs, implements UCI protocol and requires a chess GUI. Popular GUIs are:

Download Lc0

The latest stable version of Lc0 is v0.31.2.

Older versions and beta releases are available at our GitHub releases page.

Windows

Every package contains:

lc0.exe (the engine),
791556.pb.gz or 753723.pb.gz (a default network file), and
lc0-training-client.exe (only needed if you would like to contribute training games).

Depending on which hardware you have, different version of Lc0 will be best for you.
Pick the row that matches your hardware from the following table:

Setting up Lc0 in chess GUI

This section contains instructions to add Lc0 in various chess GUIs.

Configuration

For most of the use cases, default configuration of Lc0 is good for use. If you have NVidia RTX GPU, you may want to change the default network to one of the larger ones.

Hardware for Lc0

TODO(Write which hardware to buy for different usage and budget, from consumer to top-level-possible)

Benchmark

TODO(Write about our benchmark spreadsheet and how to contribute to it. If instrucions are long enough, maybe factor out in a separate document)

The easiest way to play Lc0 if you don’t want to install it on your computer is to use one of Lichess bots. You can find them in the respective lichess team. The ones that are almost always online are Lc0Test, leela2200 or MiniHuman. There is also an official Leela bot called LeelaChess, but most of the time it’s offline.
The Maia bots run on lc0 networks as well, you can play them on lichess. A new knight odds bot called LeelaKnightOdds is now accepting challenges on lichess. Discover more about the bot and learn how to challenge it by visiting this link.

Troubleshoot

Running out of memory

Unlike traditional A/B chess engines, Lc0 has to keep entire search tree in memory. That means that as the search goes, memory consumption increases. Lc0 requires roughly 250-350 bytes of memory per node, that means that typical RAM sizes can only handle so many nodes:

Play