Run the following command to build the docker for tensorcircuit at parent path:

sudo docker build . -f docker/Dockerfile -t tensorcircuit

One can also pull the official image from DockerHub as

sudo docker pull tensorcircuit/tensorcircuit

Run the docker container by the following command:

sudo docker run -it --network host --gpus all tensorcircuit

# if one also wants to mount local source code, also add args `-v "$(pwd)":/app`

# using tensorcircuit/tensorcircuit to run the prebuild docker image from dockerhub

# for old dockerfile with no runtime env setting
# sudo docker run -it --network host -e LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.0/targets/x86_64-linux/lib -e PYTHONPATH=/app -v "$(pwd)":/app  --gpus all tensorcircuit

export TF_CPP_MIN_LOG_LEVEL=3 maybe necessary since jax suprisingly frequently complain about ptxas version problem. And export CUDA_VISIBLE_DEVICES=-1 if you want to test only on CPU.

The built docker has no tensorcircuit pip package installed but left with a tensorcircuit source code dir. So one can python setup.py develop to install tensorcircuit locally (one can also mount the tensorcircuit codebase on host) or pip install tensorcircuit within the running docker.