katago.md

Getting Started with Katago

KataGo is an open-source AI, based in part on DeepMind's AlphaGo. We use KataGo for various analysis-related repositories, and so if you're working with Otrego libaries, you may need to install and run KataGo.

Installing KataGo

KataGo can be installed from he KataGo release page or from a package manager like Homebrew (OSX) or apt (Ubuntu).

KataGo works with either CUDA or OpenCL. With OpenCL, you can use it without needing a GPU.

In addition to the built-binaries, KataGo needs 3 configuration files to run:

1. Model: You can get that from the KataGo releases

• Note: There are a number of different models. If you want some nets that are much faster to run, try any of the "b10c128" or "b15c192" Extended Training Nets here which are 10 block and 15 block networks from earlier in the run that are much weaker but still pro-level-and-beyond. https://d3dndmfyhecmj0.cloudfront.net/g170/neuralnets/g170-b10c128-s197428736-d67404019.bin.gz

2. GTP Config: You can get that from the KataGo releases
3. Tuning Parameters. This is set via running katago benchmark -tune.`

• OSX
  • OSX generally comes pre-installed with OpenCL
  • Run brew install katago
  • Download the models. Brew comes with a models, which you can get with
    • KATAGO_GTP_CONFIG=$(brew list --verbose katago | grep gtp)
    • KATAGO_MODEL_PATH= $(brew list --verbose katago | grep .gz | head -1)
  • Tune Katago: katago benchmark -tune -config $(KATAGO_GTP_CONFIG) -model $(KATAGO_MODEL_PATH)
    • This will output configuration to $HOME/.katago
  • KataGo should now be operational!

Using KataGo via UI

• OSX/Linux
  • If you want to try it out with a UI, download Lizzie and change the engine command to: /path/to/katago gtp -config /path/to/config.cfg -model /path/to/model.gz
    • I had better luck editing the config.txt file that comes with Lizzie directly, rather than trying to set the engine in the Java UI.