This repository provides a dockerfile for building a runtime environment for KataGo v1.11.0.
日本語の説明はこちらを参照ください。
The following are additions or limitations to the requirements of KataGo.
Learn more about how to install docker and nvidia-container-toolkit
Clone this repository:
$ git clone https://github.com/GrahamML/docker-for-KataGo.git
$ cd ./docker-for-KataGo/dockerfile
$ docker build --tag=['image_name:tag'] .
katago-v1.11.0-cuda11.1-linux-x64.zipg170e-b20c256x2-s5303129600-d1228401921.bin.gzg170-b30c320x2-s4824661760-d1229536699.bin.gzg170-b40c256x2-s5095420928-d1229425124.bin.gzdocker command to sudo docker. $ docker build --tag=katago:latest .
The following shows the version of Ubuntu, CUDA, and cuDNN in this container.
| Ubuntu | CUDA Toolkit | cuDNN |
|---|---|---|
| 18.04 | 11.1.1 | 8.0.5 |
Start the docker image as follows:
$ docker run \
-it \
--net host \
--runtime nvidia \
--entrypoint /bin/bash \
--name [container_name] \
[image_name:tag]
nvidia-docker2 case. If it’s nvidia-container-toolkit case, change it appropriately.Launch the KataGo in this container. The following is an example of launching in benchmark mode and GTP mode.
$ cd katago
$ ./katago benchmark \
-model g170-b40c256x2-s5095420928-d1229425124.bin.gz
$ cd katago
$ ./katago gtp \
-model g170-b30c320x2-s4824661760-d1229536699.bin.gz \
-config default_gtp.cfg
rules = tromp-taylor of /workspace/katago/default_gtp.cfg in the current container.You can immediately contribute to “Katago Distributed Training” with this Docker image.
See the following steps.
./katago/contribute_example.cfg file
contribution option as follows:
$ ./katago contribute -config contribute_example.cfg
See this wiki.
