SageMaker

[JohnTigue]

See also #50.

Your own Stable Diffusion endpoint with AWS SageMaker

[JohnTigue]

Expensive Cat pictures with Stable Diffusion on AWS SageMaker in CDK

[JohnTigue]

Keep finger crossed. Awesome contributors of Automatic1111 WebUI are on the way to make Dreambooth on 8-10 GB VRam possible:

AWS Sagemaker Studio is getting a free tier called Lab specifically for deep learning training. 8 hours of T4 per day (2x 4 hours), that's about 120 USD in compute monthly (4 USD per day) for free. It uses a Jupyter interface like Google Colab.

[JohnTigue]

Here's a guy who deploys A1111 to EC2 and then in his todo notes says, ~"migrate to SageMaker": https://github.com/ManyHands/hypnowerk/issues/5#issuecomment-1381199618.

SageMaker? Sure good idea: could run the Colab notebooks privately on good GPUs. But A1111 in SageMaker? Do tell… go on…

[JohnTigue]

Running the webui on SageMaker Studio Lab #4820: "Jupyter notebook for running stable-diffusion-webui on (free) SageMaker Studio Lab:"

[JohnTigue]

Here's more A1111 on SageMaker.

[JohnTigue]

This is nice work Stable Diffusion WebUI (AUTOMATIC1111) for SageMaker Studio Lab:

This notebook will install and run the stable-diffusion-webui repository by AUTOMATIC1111 on SageMaker Studio Lab

With the default model you can run textual inversion training, hypernetwork training and image generation, but no native training

[JohnTigue]

We'll want to have Docker images with SD on them, made to work with SageMaker: Using Docker containers with SageMaker. I bet it would be pretty easy to get Invoke and JupyterLab running on the same SageMaker instance.

[JohnTigue]

We can use our foo@manyhands.io accounts to sign into SageMaker, see Identity and Access Management for Amazon SageMaker:

You can sign in to AWS as a federated identity by using credentials provided through an identity source. AWS IAM Identity Center (successor to AWS Single Sign-On) (IAM Identity Center) users, your company's single sign-on authentication, and your Google or Facebook credentials are examples of federated identities. When you sign in as a federated identity, your administrator previously set up identity federation using IAM roles. When you access AWS by using federation, you are indirectly assuming a role.

[JohnTigue]

They seem to have SD enabled instances ready to roll as of November: Generate images from text with the stable diffusion model on Amazon SageMaker JumpStart. Note that this does not include an SD webUI such as Auto1111 or InvokeAI.

[JohnTigue]

Train and deploy a custom GPU-supported ML model on Amazon SageMaker

[JohnTigue]

SD txt2img on CLI inside SageMaker:

• Generate images from text with the stable diffusion model on Amazon SageMaker JumpStart
• 2022-11 what's new: SageMaker JumpStart now provides Stable Diffusion and Bloom models

[JohnTigue]

SIWG in SageMaker: Identity and Access Management for Amazon SageMaker:

You can sign in to AWS as a federated identity by using credentials provided through an identity source. AWS IAM Identity Center (successor to AWS Single Sign-On) (IAM Identity Center) users, your company's single sign-on authentication, and your Google or Facebook credentials are examples of federated identities.

[JohnTigue]

Generate images from text with the stable diffusion model on Amazon SageMaker JumpStart

[JohnTigue]

• Amazon SageMaker Pricing
• Available Studio Instance Types