Skip to content

Docker Installation

Deploy PSuM with Docker

Before starting, make sure Docker is installed. For Windows users, we recommend installing WSL2 first, then Docker Desktop.

Docker Official Site

no-gpu Deployment

Suitable for users without a GPU or who only need logic testing. Navigate to the project root directory and enter the following commands in the terminal to disable CUDA:

rm -f config.mk.local
echo "USE_CUDA := 0" > config.mk.local
Enter the following in the terminal:
# 1. Pull the image
sudo docker pull crpi-vy93epqqxgjguf3h.cn-shanghai.personal.cr.aliyuncs.com/psum_env_docker/docker_images:no_gpu
# 2. Run the container (execute in the psum project root directory)
sudo docker run -it --name psum_dev -v $(pwd):/workspace crpi-vy93epqqxgjguf3h.cn-shanghai.personal.cr.aliyuncs.com/psum_env_docker/docker_images:no_gpu bash
# 3. First compilation inside the container; you should see the build succeed.
./build.sh
# 4. To re-enter the container each time you need to use it:
sudo docker start psum_dev # If the container is not running, use start to launch it
sudo docker exec -it psum_dev bash # Enter the container
If you need to rebuild the Docker image from scratch, refer to jump to no-gpu.

Deploy using cuda-12.4 Docker image

Prerequisites:

  • NVIDIA driver installed on the new machine (supporting CUDA 12.4)
  • Docker installed on the new machine
  • NVIDIA Container Toolkit installed on the new machine (required for --gpus all; WSL2 users can skip this step)

You may need to create a config.mk.local file in the project root directory and configure the SM version.

rm -f config.mk.local
echo "CUDA_ARCH := sm_89" >> config.mk.local # Nvidia RTX 20 series: sm75 (default, can skip this step). RTX 30 series: sm86. RTX 40/50 series: sm89;
Enter the following in the terminal:
# 1. Pull the image
sudo docker pull crpi-vy93epqqxgjguf3h.cn-shanghai.personal.cr.aliyuncs.com/psum_env_docker/docker_images:cuda_12_4
# 2. Run the container (execute in the psum project root directory)
sudo docker run -it --gpus all --name psum_dev -v $(pwd):/workspace crpi-vy93epqqxgjguf3h.cn-shanghai.personal.cr.aliyuncs.com/psum_env_docker/docker_images:cuda_12_4 bash
# 3. First compilation inside the container; you should see the build succeed.
./build.sh
# 4. To re-enter the container each time you need to use it:
sudo docker start psum_dev # If the container is not running, use start to launch it
sudo docker exec -it psum_dev bash # Enter the container
If you need to rebuild the Docker image from scratch, refer to jump to nvidia-sm75.