Docker container? about ai-horde

[wolfgangmeyers]

docker image probably shouldn't include models, since there are over 100 of them now. But the worker needs a configuration to automatically download configured models without prompting the user.

[db0]

I could, if someone makes one :D

[manioo8]

@helloeverything1 in the steps for setting up the worker you have to download a 4GB model. what approach would be better for you, having this model included in a docker container, having the container download the model from a slow mirror during the first run, or downloading the model yourself from https://huggingface.co/CompVis/stable-diffusion-v-1-4-original and providing it to the docker image?

[helloeverything1]

I sorta just saw a cool open source project and wanted to help, not sure if I am the best one to make this decision. But I feel like having the container download the model on its first run is probably best for me because its self contained but the image doesn't have to be huge. (IDK if it matters at all for AI but with a mirror files could be updated on its side and the container could update itself)

[0xdevalias]

My /2c is that it would kind of depend on the use case of the Dockerfile. I think having a more lightweight one that doesn't include the model directly in it's layers would make a lot of sense for some use cases, but I could see how having the model included and cached within the Docker container's layers could make sense in other use cases.