Garden: tools to simplify access to scientific AI advances.
Project description
🌱 Garden: FAIR AI/ML Model Publishing Framework
At a Glance:
- Easy Model Publishing: Publish pre-trained AI/ML models from a notebook with just a few commands
- Reproducible Environments: Use containers to ensure consistent execution across different systems
- Remote Execution: Run your models (or others) remotely on HPC resources seamlessly
- Discoverable Collections: Organize models into "Gardens" for easy discovery and comparison
- Metadata Management: Capture and manage metadata of related datasets, papers, or code repositories for better searchability
Why Garden?
Garden addresses key challenges faced by academic researchers in discovering, reproducing, and running AI/ML models:
-
Reproducibility: Garden eliminates environment inconsistencies by containerizing models, ensuring they run consistently across different systems.
-
Discoverability: With curated "Gardens" of models, researchers can easily find, compare, and curate relevant models for their work.
-
Accessibility: Garden simplifies the process of running models on diverse computing resources, from local machines to HPC clusters, via Globus Compute integration.
-
Time-saving: By handling environment management and system-specific quirks, Garden significantly reduces the time researchers spend on setup and configuration.
-
Collaboration: FAIR principles (Findable, Accessible, Interoperable, Reusable) and standardized publishing make it easier for researchers to share their work and build upon others' contributions.
Garden aims to let researchers focus on their science, not on the intricacies of software environments and computing infrastructure.
What's a Garden?
A "Garden" is a citable collection of published pre-trained AI/ML models, called "Entrypoints".
Ok, What's an Entrypoint?
An "Entrypoint" is just a python function you define in a regular jupyter notebook which typically invokes one or more of your pre-trained models.
When you give us that notebook, we "freeze it in amber" by containerizing it (along with any environment dependencies) and give you a citable DOI in return.
Now, anyone with the DOI can easily invoke that exact function, in the exact same environment, on any remote compute resources they have access to (via Globus Compute):
Quick Start
-
Install the garden CLI:
pipx install garden-ai
-
Set up Docker on your system (required for local development and testing)
We recommend installing Docker Desktop for most users.
-
Start a notebook in an isolated environment:
garden-ai notebook start my_model.ipynb --base-image=3.10-sklearn
-
Define a function invoking your model in the notebook and publish it:
garden-ai notebook publish my_model.ipynb
For a more detailed walkthrough, check out our 15-minute tutorial.
Documentation
For more documentation, including installation instructions, tutorials, and API references, see our latest docs.
Contributing
We welcome contributions from the community! Please see our Contributing Guide for more information on how to get started.
Support
This work was supported by the National Science Foundation under NSF Award Number: 2209892 "Frameworks: Garden: A FAIR Framework for Publishing and Applying AI Models for Translational Research in Science, Engineering, Education, and Industry".
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for garden_ai-2.2.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | cafd4f41495dd161824aa9ce79958b06feb111dd2f2a48df1702bc8ef62eb536 |
|
MD5 | 34b65b7eab3ba376cbfd954c6cf4c551 |
|
BLAKE2b-256 | bedc42f4814c32a1ecb43be238477a5cb4d1cf676a750e80145222f670d234a1 |