[Moreh] Test alpha-zero-general on Nvidia-A100 machine

Prepare

Data

• Don't have to prepare data as they will be generated during training

Code

git clone https://github.com/suragnair/alpha-zero-general
cd alpha-zero-general

Environment

On Nvidia-A100 machine, NVCC version 11.2.0

conda create -n alpha-zero-general python=3.8
conda activate swinbert

Run

1. Reproduce the CUDNN_STATUS_EXCECUTION_FAILED error

First, install the CUDA container Toolkit https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html. Change the directory to alpha-zero-general. Then run the below script to create container environment.

./setup_env.sh

If sudo is needed to run docker, run

sudo /bin/bash setup_env.sh

Finally, run

docker exec -ti pytorch_notebook python main.py

to reproduce the error.