This repository is obsolete. ShivamShrirao made a better implementation of this.
Here is his repository: https://github.com/ShivamShrirao/diffusers.
notes
the original source is from https://github.com/ShivamShrirao/diffusers
I used this revision of the xformers library pip install git+https://github.com/facebookresearch/xformers@1d31a3a#egg=xformers
Tested on a nvidia 3060 12G.
My machine:
debian linux,
64GB of ram,
a nvidia 3060 as the main gpu for the desktop,
a secondary nvidia 1060 6G.
DreamBooth training example
open comand.sh and edit the folder names and the relevant parameters
DREAMBOOTH_SECONDARY is the device where you want to put the vae.
If unsure, leave "cpu". If you have a secondary GPU, then use "cuda:1". If you want to use the main gpu, then "cuda"
The EFFICIENT_TRAINER is set to 1 to use the most efficient setup I found.
For a bit more precise training use 0.
PARALLEL_TRAINING. If 0, the execution is alternating between cpu and gpu
If set to 1, the gpu and secondary device will run in parallel.
Parallel training doesn't increase VRAM.