The AWS EC2 instance you would want to use depends on the model you want to finetune. In this article, we’ll detail what EC2 instance you should use for your specific model.

​
Dreambooth

Dreambooth requires about 24GB of RAM, which means that A10s work well for this task. AWS offers A10 GPUs with their G5 instances. We recommend you use this G5 specification:

​
AWS EC2 G5.2xlarge

  • 100GB storage

  • Amazon Linux 2

Step 1. Install Docker

$ sudo yum update -y
$ sudo amazon-linux-extras install docker
$ sudo service docker start
$ sudo usermod -a -G docker ec2-user

Step 2. Install Rubbrband

$ sudo su
$ pip install rubbrband

Step 3. Start a Sample training run

This is from our quick start guide. It’ll download some sample data, and start training Dreambooth.

git clone https://github.com/rubbrband/sample_dataset.git
git clone https://github.com/JoePenna/Stable-Diffusion-Regularization-Images.git --depth 1
mkdir regDir
mv ./Stable-Diffusion-Regularization-Images/man_unsplash ./regDir/man

# start training
rubbrband train dreambooth --class-word man --dataset-dir ./sample_dataset --reg-dir ./regDir --log-dir ./logs

​
ControlNet

Docs work in progress.