/tutorials

Tutorials and usage examples

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

🏃‍♀️Runhouse 🏠
Walkthrough

These tutorials introduce you to the tools and usage patterns of Runhouse. We've devised them to chart a fun path through our features, but you're welcome to hop around if you prefer.

  1. 🐣 Runhouse Basics - Fun with Stable Diffusion and FLAN-T5
  2. 🧑‍🎨 Fancy Runhouse - Dreambooth in <10 Minutes
  3. 👩‍🚀 Portability - DALL-E to SD img2img from Notebook to Inference Service
  4. 👩‍👩‍👧‍👧 [WIP] Distributed - Large Model Training and Inference
  5. 👩‍🔧 [WIP] Pipelines - Fine-tuning BERT and Deploying

If you would be so kind, we would love if you could have a notes doc open as you install and try Runhouse for the first time. Your first impressions, pain points, and highlights are very valuable to us.

🛫 Installation & Setup

See getting started.

tldr;

pip install runhouse
sky check
# Optionally, for portability (e.g. Colab):
runhouse login

⏰ If you only have 10 minutes:

  • Take a look at the Stable Diffusion example to understand how Runhouse allows you to interact with remote compute.
    • The FLAN-T5-XL example then shows how we can easily reuse hardware and services.
  • See our dreambooth tutorials (training, inference). We think they're the easiest way anywhere to run dreambooth on your own cloud GPU (for managed dreambooth, check out Modal Labs's dreambooth or StableBoost).
  • See how to launch a Gradio app to run CLIP Interrogator (generate Stable Diffusion prompts from images), here.
  • See our BERT fine-tuning pipeline example, here.

🕳 If you want to go deeper:

Please check out our docs site! In addition to documentation, we also provide both high level and detailed overviews of the Runhouse architecture.

🚨 This is an Alpha 🚨

Runhouse is heavily under development, and we expect to iterate on the APIs before reaching beta (version 0.1.0).

🙋‍♂️ Getting Help

Please join our discord server here to message us, or email us (donny at run.house or josh at run.house), or create an issue.

🕵️‍♀️ Where is the compute?

Runhouse is not managed compute or data. All of the compute and data in Runhouse lives within your own infra and cloud accounts. As such, you'll need credentials with at least one of AWS, GCP, or Azure (and Lambda Labs and IBM coming soon) to try these tutorials, as well as quota approval for GPU resources (See here for more on this). If you're looking for a managed compute experience without a cloud account, we'd recommend our friends at Modal Labs or Anyscale. At some point we plan to support them as compute providers in Runhouse as well. Other sources of compute, such as on-prem or Kubernetes, are also on the roadmap (likely through our friends at SkyPilot).

👷‍♀️ Contributing

We welcome contributions! Please check out out contributing if you're interested.