Installation using Docker

[rebecca-lay3rs]

inside the repo we can find Dockerfile

Dockerfile sets the total machine environment for the repository

To build the Dockerfile, we use the Makefile with all the shell commands (edited from the original one)

Make commands

make build-image
make run

inside Makefile we have two functions:

• build-image (for Docker image) and run (run Container)
• they are separate from each other: if you need to change the configurations of the env you need to change the build-image, if you just want to change how the Container is launched, just change the run command

Docker commands

docker ps
docker ps -a
docker stop ID
docker images
docker image rm ID
docker exec -it gsa /bin/bash

The Docker run command is configured to mount the home/appuser/working_dir/ to the github cloned repository ~/Grounded-Segment-Anything, so we can change the files inside the Container (/home/appuser/working_dir) and it will change the local hosto folder ~/Grounded-Segment-Anything. This is important because the Docker Container can crash and we might lose the work.

Tmux inside Docker https://stackoverflow.com/questions/51809181/how-to-run-tmux-inside-a-docker-container

• need to add to Dockerfile

apt-get update && \
  apt install -y tmux &&

Jupyter Notebook

• we need to install jupyter inside Container: pip install notebook

• the goal is to launch jupyter and the calculations from inside the Docker Container (a bit like Conda env) so we need to run the Docker Container already bind to the port that we want to bind the jupyter from inside the container, such as: (https://stackoverflow.com/questions/38830610/access-jupyter-notebook-running-on-docker-container)

  Host machine (the VM we use): docker run -it -p 8080:8080 image:version
  
  Inside the Container : jupyter notebook --ip=* --no-browser --port 8080 --allow-root
  
  Host machine access this url : localhost:8888/tree‌

H100 Compute Capability

• the Dockerfile specifies that the max CC is sm_86 (8.6) but our H100 is sm_90 (9.0)
• TODO: try in A100 Azure

https://developer.nvidia.com/cuda-gpus GPU	Compute Capability
NVIDIA H100	9.0
NVIDIA A100	8.0

[rebecca-lay3rs]

(base) root@h100-rebecca:~/Grounded-Segment-Anything# docker ps
CONTAINER ID   IMAGE     COMMAND                  CREATED       STATUS       PORTS                                       NAMES
b464b2bb26c4   gsa:v0    "/opt/nvidia/nvidia_…"   2 hours ago   Up 2 hours   0.0.0.0:8080->8080/tcp, :::8080->8080/tcp   gsa
(base) root@h100-rebecca:~/Grounded-Segment-Anything# docker stop b464b2bb26c4
b464b2bb26c4
(base) root@h100-rebecca:~/Grounded-Segment-Anything# docker images
REPOSITORY   TAG       IMAGE ID       CREATED       SIZE
gsa          v0        813eb17e9d25   3 hours ago   22.5GB
(base) root@h100-rebecca:~/Grounded-Segment-Anything#