README.md

---
title: README  
emoji: 📚  
colorFrom: pink  
colorTo: gray  
sdk: static  
pinned: false  
---

# 📚 BigLAM

A community-run home for machine-learning-ready datasets from libraries, archives, and museums.

Most cultural-heritage data wasn't originally prepared with ML workflows in mind — it lives in catalogue systems, IIIF endpoints, METS/MODS records, and various idiosyncratic formats that each institution has its own version of. BigLAM is a place where those datasets get repackaged into formats ML practitioners can actually load and work with, contributed by the people who know the source material best.

The org started as a [datasets hackathon](https://github.com/bigscience-workshop/lam) inside the [BigScience](https://bigscience.huggingface.co/) project in 2022 and has grown into a standing community for cultural-heritage ML.

## What's here

The org is datasets-first: 46+ image, text, and tabular collections from libraries, archives, and museums, prepared so they load cleanly with the `datasets` library. A handful of [models](https://huggingface.co/biglam?other=model) and [spaces](https://huggingface.co/biglam?other=space) live here too — mostly early experiments from the BigScience-era hackathon.

For task-specific, deployable models built on top of these datasets, see the sibling org [small-models-for-glam](https://huggingface.co/small-models-for-glam).

## Contributing a dataset

If you've prepared a LAM dataset that other researchers might use, the best home is usually your **institution's own Hugging Face organisation** (e.g. [`NationalLibraryOfScotland`](https://huggingface.co/NationalLibraryOfScotland)). Institutional ownership signals authority over the data and makes long-term maintenance easier. Setting up a new org on the Hub is [free and quick](https://huggingface.co/organizations/new).

If your institution isn't on the Hub yet, or you'd prefer to host the dataset here, [open a discussion](https://huggingface.co/spaces/biglam/README/discussions) and we'll help get it set up under BigLAM. Useful additions are typically datasets where the format conversion (METS/ALTO → parquet, IIIF manifest → loadable image splits, etc.) has already been done and the licensing is clear enough for open release.

**Already have a dataset here that should sit under your institution's org?** Open a discussion or issue on the dataset repo — we're happy to transfer ownership.

---

60+ contributors over the years. Day-to-day maintenance is light-touch; for help with a contribution, open a discussion and someone will see it.