This repository combines the Nibbler chess GUI with Leela Chess Zero, the UCI-compliant neural network chess engine, as prebuilt binaries for MacOS.

@fohristiwhirl is the creator of the Nibbler chess GUI. Without a MacOS system, it's challenging to integrate Nibbler, with the Electron software framework for MacOS. Until he figured this out and provides a MacOS build of Nibbler, you can use this Nibbler-for-macOS release. It contains

• Nibbler Chess GUI from https://github.com/fohristiwhirl/nibbler

• Electron Framework from https://github.com/electron/electron

• Chess engine Lc0 from https://github.com/LeelaChessZero/lc0 compiled with clang 9.0 for OpenCL and BLAS (Apple vecLib) backends

The chess engine Lc0 needs a weights file (network).

• download the release file (not the source code) from https://github.com/twoplan/Nibbler-for-macOS/releases
• unpack this zip-file to a directory of your choice (you probably just did this)
• start Nibbler.app (use right click and open)
• from the menu entry Engine select your path to a (previous downloaded) weights file and the lc0 chess engine (inside this folder, named lc0)
• choose a backend: blas (uses cpu) or opencl (uses gpu)
• you are ready now; take a look at menu entry Electron/About

You can help yourself:

• download and unpack the latest source code of Nibbler from https://github.com/fohristiwhirl/nibbler/releases
• replace all files in Nibbler.app/Contents/Resources/app/ with them

• the first time a new network size is selected for backend opencl, leelaz_opencl_tuning gets created or updated. Depending on your gpu, this can take some minutes.
• use the opencl backend on MacBooks with care. It could overheat your system. Check this with Intel power gadget https://software.intel.com/en-us/articles/intel-power-gadget
• all settings can manually be edited; they are saved in ~/Library/Application Support/Nibbler/config.json
• got these nps values with blas and opencl for different network sizes:

	size 320x24: (60130, go nodes 1000)	    38 nps (blas) |  42 nps (opencl)
	size 256x20: (42850, go nodes 1000)	    37 nps (blas) |  22 nps (opencl)	
	size 128x10: (56170, go nodes 10000)	   383 nps (blas) | 406 nps (opencl)	
	11258-48x5-se: (go nodes 10000)	          2730 nps (blas) | 175 nps (opencl)

• https://lczero.org/networks/
• https://github.com/dkappe/leela-chess-weights/wiki/Bad-Gyal
• https://github.com/dkappe/leela-chess-weights/wiki/Distilled-Networks
• https://github.com/jhorthos/lczero-training/wiki/Leela-Training