Introduction

Foundry

Foundry-ML is a Python library that simplifies access to machine learning-ready datasets in materials science and chemistry.

Features

• Search & Discover - Find datasets by keyword or browse the catalog

• Rich Metadata - Understand datasets before downloading with detailed schemas

• Easy Loading - Get data in Python, PyTorch, or TensorFlow format

• Automatic Caching - Fast subsequent access after first download

• Publishing - Share your own datasets with the community

• AI Integration - MCP server for AI assistant access

• CLI - Terminal-based workflows

Quick Example

from foundry import Foundry

# Connect
f = Foundry()

# Search for datasets
results = f.search("band gap", limit=5)

# Load a dataset
dataset = results.iloc[0].FoundryDataset
X, y = dataset.get_as_dict()['train']

# Get citation for your paper
print(dataset.get_citation())

Installation

pip install foundry-ml

For cloud environments (Colab, remote Jupyter):

f = Foundry(no_browser=True, no_local_server=True)

What's Next?

Getting Started Installation Quick Start	User Guide Searching Loading Data ML Frameworks	Features CLI MCP Server HuggingFace

Project Support

This work was supported by the National Science Foundation under NSF Award Number: 1931306 "Collaborative Research: Framework: Machine Learning Materials Innovation Infrastructure".

Foundry brings together components from:

• Materials Data Facility (MDF)

• Data and Learning Hub for Science (DLHub)

• MAST-ML