DMI Dockerized Services

This repo contains a collection of Dockerized services allowing them to be run via a command line inside a controlled environment. Each container should contain a Dockerfile to build the proper environment and, possibly, a docker-entrypoint.sh file running a default commnads. It also may contain a command line interface file allowing different arguments to be passed to the service.

These Services were designed primarily to work with our DMI Service Manager which allows these services to be run via an API and can be integrated with our tools such as 4CAT. The services can also be run independently on local files and some contain their own APIs. Explore their individual README.md files for more information.

Services

Service Name	Folder	Source	Notes
Whisper	openai_whipser	https://github.com/openai/whisper
CLIP	openai_clip	https://github.com/openai/CLIP/
BLIP2	blip2	https://huggingface.co/Salesforce/blip2-opt-2.7b
Stable Diffusion	stable_diffusion	https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
PixPlot	DMI PixPlot repo	https://dhlab.yale.edu/projects/pixplot/	The DMI PixPlot was developed prior to the DMI Service Manager as a stand alone service, but it now also works with the DMI SM. You can find instructions here on installation and setup.

Installation

1. Install Docker itself

• You can find information here to install on Windows, Mac, or Linux.

2. Build the Docker images for your desired services

Building a Docker image

1. Clone this repository

• git clone https://github.com/digitalmethodsinitiative/dmi_dockerized_services.git

2. Navigate to the desired image

• e.g., cd dmi_dockerized_services/openai_whipser for our Whisper image
• Individual services may have README.md files similar to this which additional information on running the service and types of commands that you can send

3. Build the image

• docker build -t image_name .
• image_name can be whatever you wish, but you will need it to run the image and set up the DMI Service Manager (e.g., docker build -t whisper . for Whisper)

4. You can now run the image individually or set up the DMI Service Manager to run it for you and enable it in 4CAT

Enable a service in the DMI Service Manager

Complete instructions are available in the DMI Service Manager wiki

1. Enable the service in DMI Service Manager's config.yml under the DOCKER_ENDPOINTS section. E.g.,:

whisper: 
 image_name: whisper
 local: True  # Set to True if 4CAT is running locally
 remote: False  # Set to True if 4CAT is running remotely
 command: whisper
 data_path: /app/data/

Enable a service in 4CAT

1. First enable the service in the DMI Service Manager (see above)
2. On your 4CAT server, navigate to Control Panel -> Settings -> DMI Service Manager (you must be an administrator).

• Set the DMI Service Manager server/URL to the server where the DMI Service Manager is running (e.g. http://localhost:4000)
• Set DMI Services Local or Remote to either local or remote depending on whether 4CAT and the DMI Service Manager are on the same server
• Find and enable the server you have just created and adjust any relevant settings
• Run the analysis on an appropriate dataset!

Run the container persistently

Most services are set up to run persistantly allowing you to send multiple commands as desired.

Run container as daemon from image

docker run -v $(pwd)/data/:/app/data/ --name container_name --gpus all -d image_name

• -v $(pwd)/data/:/app/data/ mounts the data directory in your current working directory to the container
• --name container_name names the container "container_name"
• --gpus all is needed for the container to use the host GPUs
• -d runs the container and disconnects
• the final argument is the tag assigned to your image when built above

Connect to container to run container_name commands

docker exec -it container_name bash

Run the container for a single use

This docker run command combines the above to create a one time use Docker container to run commands

Assuming you have a folder data in your current working directory with audio files in it, run: docker run --rm -v $(pwd)/data/:/app/data/ --gpus all container_name command_of_choice

• --rm removes the container after it has completed