# Multimodal Universe: DESI Early Data Release SV3 Spectra

## Facts

- Provider: Multimodal Universe Collaboration (Polymathic AI, UniverseTBD)
- License: CC-BY-4.0 (https://data.desi.lbl.gov/doc/acknowledgments/)
- Size: 634 files
- Format: PNG, Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/UniverseTBD/mmu_desi_edr_sv3
- Access: public, readable without an account
- Category: Astronomy & Space
- Source: https://arxiv.org/abs/2412.02527
- Page: https://scigantic.com/archive/4e33cf76-f213-5f1e-b484-dddd8e13607f
- Keywords: Astronomy & Space, spectroscopy, PNG, Parquet, Hugging Face

## About

Optical spectra from the Dark Energy Spectroscopic Instrument (DESI) Early Data Release (EDR), Survey Validation 3 (the One-Percent Survey), packaged as part of the Multimodal Universe collection of machine-learning-ready astronomical data. The modality is 1D spectroscopy: each of 1,126,441 objects has a spectrum of 7,781 pixels starting at 3,600 Angstrom with flux, inverse variance, line spread function width and mask arrays, plus the DESI redshift Z, ZERR and ZWARN flag, Galactic extinction and Legacy Surveys g, r and z total and fiber fluxes. The README lists 20 columns in 306 HEALPix partitions and 61.8 GiB on disk. Listing the repository (measured 2026-11-13) shows the main catalog folder mmu_desi_edr_sv3/ (dataset/, hats.properties, partition_info.csv, point_map.fits), the margin catalog mmu_desi_edr_sv3_10arcs/ and collection.properties; about 69 GB in all. Data are a HATS catalog: Apache Parquet partitioned by HEALPix tile, with a 10 arcsecond margin catalog for cross-matching. Open with LSDB (lsdb.open_catalog on the Hugging Face URL, with cone searches and crossmatch), or read the Parquet with pandas, pyarrow, dask or DuckDB. Uses include redshift estimation, spectral classification and cross-modal learning with the Legacy Surveys images. Part of the Multimodal Universe (arXiv:2412.02527); released under CC-BY-4.0, cite the Multimodal Universe paper and the original survey.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("UniverseTBD/mmu_desi_edr_sv3")
print(ds)
```

## Data structure

File types in sampled files: PNG, Parquet

Example file: `mmu_desi_edr_sv3/dataset/Norder=10/Dir=2500000/Npix=2506751.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/4e33cf76-f213-5f1e-b484-dddd8e13607f
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/4e33cf76-f213-5f1e-b484-dddd8e13607f
- Over MCP: `get_schema_card` with id `4e33cf76-f213-5f1e-b484-dddd8e13607f` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Astronomy & Space datasets](https://scigantic.com/topics/astronomy-and-space)
- [Spectroscopy datasets](https://scigantic.com/topics/spectroscopy)
- [PNG datasets](https://scigantic.com/topics/png)
