Skip to content

Running lc0 on Android with a chess GUI

Leandro Álvarez González edited this page Mar 2, 2020 · 85 revisions

Follow these simple steps and you'll be running lc0 on your Android device. No root needed. Just the right engine, a weights file and a chess GUI.

Video Tutorial

Here's a small video showing the steps described below, just in case you need to see it in action:

https://streamable.com/fozkg

Get lc0 for Android

Supported Devices

There are currently two architectures supported: armv7-a and aarch64. In general older devices are 32-bit in which case you must download the armv7-a builds. The more recent devices are 64-bit so if you have one you should get the aarch64 builds that are much faster. If you are not sure which architecture to choose, use this app to lookup the architecture of your Android device.

Supported Android versions

All the 64-bit builds on this page are compiled targeting the API Level 21 (equivalent to Android 5.0 Lollipop). The 32-bit ones target the API Level 16 (Android 4.1 Jelly Bean). Older Android versions may not work.

Which back-end to choose?

It's better to try either Eigen or OpenBLAS links first as they're easier to use and they should be faster too. As for the OpenCL ones, they are heavier and have some extra requirements:

  • Device with OpenCL support (builds are linked to OpenCL version 2.0)
  • Expect a very long delay on the first start of the engine and when a different network size is tried. This happens because the OpenCL tuner needs to optimize its parameters to your device.

Official builds for the release candidate 0.24rc2

Architecture Eigen OpenBLAS OpenCL
armv7-a - link -
aarch64 - link -

Unofficial builds for version 0.23

Architecture Eigen OpenBLAS OpenCL
armv7-a link link link
aarch64 link link link

Unofficial builds for version 0.22

Architecture Eigen OpenBLAS OpenCL
armv7-a link link
aarch64 link link link

Unofficial builds with embedded weights

These are special builds for those who couldn't or wouldn't fiddle with UCI Options and full paths, or maybe those using a GUI without that capability. These builds contain a small weights file embedded into the engine itself. You'll still be able to use other weights with them, but the embedded weights will load when you didn't specify one.

Architecture Eigen OpenBLAS OpenCL
armv7-a with LD2 link link
aarch64 with LD2 link link link

Unofficial and older development builds

Architecture Eigen OpenBLAS OpenCL
armv7-a link link
armv7-a with 48x5 link
aarch64 link link link
aarch64 with 48x5 link

For more information about the above builds, please browse this PR: https://github.com/LeelaChessZero/lc0/pull/848

Select a network weights file

The regular nets will work just fine, get one from the usual place:

http://lczero.org/networks

But you could choose a small network like these ones, that will be faster:

https://github.com/dkappe/leela-chess-weights/wiki/Distilled-Networks

Choose a Chess App

Any Chess App that supports external engines and setting the UCI Options should do. DroidFish and Chess for Android work quite well. Here you are the links:

https://play.google.com/store/apps/details?id=org.petero.droidfish

https://play.google.com/store/apps/details?id=com.google.android.chess

Install the GUI and copy the engine

You'll need to install DroidFish and give it access to the storage. Then unzip the lc0 engine downloaded in the first step and copy it to the DroidFish/uci folder on your Android device. After that, just open DroidFish and select the engine lc0. But don't start analysis/play yet as you'll have to configure it first.

Configure the engine

The auto-discover feature for the weights file won't work on Android, since some GUIs import the engine to a private directory. That's why you need to provide the engine with the full path to the weights file. You could use a file manager app that allows copying the full path of a file. Like this one:

https://play.google.com/store/apps/details?id=com.mi.android.globalFileexplorer

Navigate to the weights file you downloaded and copy its full path. Then paste it to the engine options, it's the first one.

If you followed until now then you should have an engine working. Start analysis and watch.

Clone this wiki locally