A Go game analysis tool for Android, providing the power of Leela Zero on your smartphone.
Leela Zero is a strong Go bot, created as an independent implementation of the Alphago Zero deep learning algorithms. It's now so strong that even on a phone, with relatively few playouts, it can achieve a high dan level.
Note that Leela Zero's analysis uses as much CPU power as it can, and will rapidly burn through battery power if not charging.
Lazy Baduk is available on Google Play. This version currently uses the last 15x192 network, from July 2018, in order to try to be as fast as practical on a phone. Larger networks take longer to initialise and to generate playouts.
Versions incorporating other networks are available for direct download:
- Latest 15x192 network d351f06e: the same as the version on Google Play
- Latest 20x256 network 33986b7f
- Latest 40x256 network 85a93684: this one especially is much larger and slower to load than the 15x192 version
- Elfv2: the ELF OpenGo version 2 weights converted for use with Leela Zero. ELF is Facebook's own implementation of a zero knowledge algorithm, and is very strong (network size 20x256).
- 9x9: uses the 9x9 weights generated by @sbf2000 here (these may not be the strongest 9x9 weights available)
- 13x13: uses the 13x13 weights generated by @alreadydone from the 19x19 weights #205 (the linked issue describes this conversion process), apparently this network is quite strong
Lazy Baduk is written in Python using Kivy. This means that it can also run on the desktop.
The app includes a Leela Zero binary compiled for Android, so all analysis is run on your device. It does not currently use the GPU, although it might be possible to compile a version supporting OpenCL for devices that support it.