Publish and Run Scientific AI Models with Garden

Run frontier models in materials science, biology, and physics with two lines of Python.

Science Needs More Than Chatbots

There are many great platforms for hosting chatbots. But you work in science, and you want the best model for predicting material tensile strength or the behavior of neutron stars. Garden is the best place to find, share, and run specialized AI models for science. (Scientific chatbots are welcome too.)

Featured Gardens

FairChem/OpenCatalyst OC20 Models

This Garden contains models trained on the OC20 dataset published by FairChem and the Open Catalyst Project. The models in this Garden are full-sized and trained on the full OC20 dataset. Both S2EF and IS2RE models ...

Materials Property Prediction with MAST-ML

Random forest models of 33 materials properties to provide predictions, error bars, and domain of applicability guidance. Models are trained and executed with the Materials Simulation Toolkit for Machine Learning (MAST-ML) from the UW-Madison Computational Materials Group. This garden also includes three batch execution variants used to screen candidate perovskites.

Conservative to Primitive Conversion in Relativistic Hydrodynamics

This garden hosts a suite of PyTorch models (and corresponding TensorRT engines) trained for conservative-to-primitive variable recovery in numerical relativity simulations, specifically tailored ...

Generative Materials Models

This Garden contains models that take free text as input and produce molecular structures as output. Models include: - Chemeleon from Hyunsoo Park and Aron Walsh at University College London - AtomGPT from Kamal Choudhary at the National Institute of Standards and Technology

Reproducible Science Needs On-Demand Models

It can take days to get another lab's model running. Models hosted on Garden are runnable in seconds, so you can build on others' work. (And you can get real users and citations for models you've developed.)

Get Started in Seconds

# "pip install garden-ai" first
from garden_ai import GardenClient

garden_client = GardenClient()
g = garden_client.get_garden("10.26311/ep98-br79")

materials = ['AgI', 'CdTe', 'BN']
# Run the model remotely and retrieve your results
result = g.predict_piezoelectric_displacement(materials)

Easy To Run, Easy To Publish

Garden uses Modal to run models in the cloud and Globus Compute to run models on Research computing clusters. Read our documentation to learn how to publish your models with Garden.

Garden AI Logo

Made Possible By:

The National Science Foundation

Award Abstract #2209892: “Frameworks: Garden: A FAIR Framework for Publishing and Applying AI Models for Translational Research in Science, Engineering, Education, and Industry”

About

Contact

GitHub

Email Us!


This project builds upon work including:

The Materials Data Facility (MDF)

NIST-supported effort to build data services to help material scientists publish and discover data.

Foundry-ML

An open source ML-ready data access tool for scientists.

Globus

Research cyberinfrastructure, developed and operated as a not-for-profit service by the University of Chicago to enable research data transfer, sharing, access, discovery, and automation.

MDF logo
Foundry logo
Globus logo