Invoke 3.0 Dockerfile #577
Conversation
InvokeAI 3.0 has a completely different structure and tries to auto-download everything. There may be better solutions for this such as skipping model downloads or similar, but after initial run the container should start quickly. Most of the symlinks should capture the models roughly into the existing structure used in stable-diffusion-webui-docker repo but further testing should be done. Docker file is based on upstream invokeai's Dockerfile
invokeai converts models into diffuser format and caches them. We don't want to have to regenerate this cache each time we rebuild container, but it will take a LOT of space on the local drive.
|
Thank you for doing this. I wanted to update to 3.0 for a while but the invoke team changed a lot and because it all supposed to be automated away it has to find how to fit the changes into the current structure. I haven't took a deeper look yet, I will come back to this later. |
| FROM alpine:3.17 as xformers | ||
| RUN apk add --no-cache aria2 | ||
| RUN aria2c -x 5 --dir / --out wheel.whl 'https://github.com/AbdBarho/stable-diffusion-webui-docker/releases/download/6.0.0/xformers-0.0.21.dev544-cp310-cp310-manylinux2014_x86_64-pytorch201.whl' | ||
| # syntax=docker/dockerfile:1.4 |
There was a problem hiding this comment.
We might need to remove this since it is not part of the official OCI image format and is incompatible with podman. or is it now?
There was a problem hiding this comment.
Are you referring to the syntax statement? I have no familiarity with podman. I intended this commit to be a jumping off point, if it helped. I am happy to dig in a bit further if needed, but I think it will be necessary to tag team it.
There was a problem hiding this comment.
yeah the syntax statement, but no worries, I want to dig deeper here anyway since a lot of stuff changed.
This should be considered a WIP, but it's an attempt at InvokeAI's v3.0 integration into the current workflow. It will need to be reviewed for conformity with @AbdBarho's standards for this repo, but I tried to symlink in directories in a way that conforms with existing structure. There may be better ways to do this, such as disabling automatic downloads and copying files around, but this is the best I could do for now, perhaps it will be useful as a starting point. It is entirely functional for me.
The Dockerfile here is based on the upstream InvokeAI's Dockerfile.
The main changes consist of pulling in the invoke repo instead of building from it, as well as symlinks to their proper place. CPU support (as you do with A1111) is easily added if docker-compose from this repo is modified to pass GPU_DRIVER.
Due to the way InvokeAI runs, it will download some "core" models and support files while providing very little debugging output. First run took over an hour of downloading upscale models and other supporting files.
Closes issue #555 #553
Update versions