https://github.com/gadicc/diffusers-docker-repro
Docker container to create reproductions of diffusers bugs
https://github.com/gadicc/diffusers-docker-repro
Last synced: 11 months ago
JSON representation
Docker container to create reproductions of diffusers bugs
- Host: GitHub
- URL: https://github.com/gadicc/diffusers-docker-repro
- Owner: gadicc
- Created: 2022-11-21T14:46:43.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-11-21T14:46:58.000Z (over 3 years ago)
- Last Synced: 2025-04-05T10:30:09.099Z (about 1 year ago)
- Language: Python
- Size: 5.18 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# diffusers-docker-repro
Copyright (c) 2022 by Gadi Cohen. MIT Licensed.
This is the docker image I've been using for debugging diffusers in a
consistent environment.
It's setup for dreambooth training can be used for anything. Will happily
accept PRs to add useful fixtures for other tests.
## Usage:
1. git clone
1. Review the `Dockerfile` but it should be good
1. Set `HF_AUTH_TOKEN` environment variable to your HuggingFace token.
1. `docker build -t diffusers --build-arg HF_AUTH_TOKEN=$HF_AUTH_TOKEN .`
1. `docker run -it diffusers`
At this point you'll get a shell inside the container and can play
around in it however you want (particularly to `git checkout`
particular commits).
## git bisect
If you're hunting for a recently introduced bug, `git bisect` is an
amazing tool. You can read up about it but in short:
```bash
$ cd .. # back to git root (from diffusers src directory)
$ git bisect
$ git good
$ git bad
# Git will now perform a binary search and drop you in the middle
# Run your test case
$ git good # if it works, OR
$ git bad # if it fails
# Repeat the procss until git gives you the answer
$ git reset # to end the session
```
If you can narrow your test case into a script that returns an error code,
you can automate the above process.