Docker image customization and minification - Githubissues

basujindal / stable-diffusion

Optimized Stable Diffusion modified to run on lower GPU VRAM

Other

3.14k stars 467 forks source link

Docker image customization and minification #223

Closed JeffersonGibin closed 1 year ago

JeffersonGibin commented 1 year ago

Docker image customization and minification

Hello everyone, I made some improvements to Dockerfile and docker-compose, I hope it will be helpful in the community.

What was improved?

Reduction in final image size from 10.3GB to 7.19GB
Update nvidia/cuda image from 11.3.1-base-ubuntu20.04 to 11.8.0-base-ubuntu22.04
Customization of the execution of files from the optimizedSD directory
Update numpy from 1.19.2 to 1.20.3

How to use execution customization?

If you need to run another file from the OptimizedSD directory, is possible to modify the default execution, by changing the environment variable APP_MAIN_FILE in docker-compose.yml and doing a new build by running:

docker-compose up --build

If you don't want to use docker-compose, you can provide the OPTIMIZED_FILE argument with the following command.

docker build --tag newimage --build-arg OPTIMIZED_FILE=optimized_img2img.py .

What was not improved?

No modifications to the project's source.

Wrong: to ensure that the tests do not have side effects, please remove container and docker images related to the project.

Ideas to improve build time in the future

The venv could be uploaded to some storage and when building the image download the environment, this will probably be considerably faster than the build step.
Positive: The image does not need to go through several steps to build the environment (compression and decompression, downloading packages)
Negative: I would need to update venv whenever new dependencies are added to the project

basujindal commented 1 year ago

Hi Jefferson, thank you for the changes, it is very helpful!