datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
FineFineWeb
FineFineWeb: A Comprehensive Study on Fine-Grained Domain Web Corpus
arXiv: Coming Soon
Project Page: Coming Soon
Blog: Coming Soon
Data Statistics
Domain (#tokens/#samples)
Iteration 1 Tokens
Iteration 2 Tokens
Iteration 3 Tokens
Total Tokens
Iteration 1 Count
Iteration 2 Count
Iteration 3 Count
Total Count
aerospace
5.77B
261.63M
309.33M
6.34B
9100000
688505
611034
10399539
agronomy
13.08B
947.41M
229.04M
14.26B
15752828
2711790
649404
19114022… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/FineFineWeb.FineFineWeb-sample
FineFineWeb: A Comprehensive Study on Fine-Grained Domain Web Corpus
arXiv: Coming Soon
Project Page: Coming Soon
Blog: Coming Soon
Data Statistics
Domain (#tokens/#samples)
Iteration 1 Tokens
Iteration 2 Tokens
Iteration 3 Tokens
Total Tokens
Iteration 1 Count
Iteration 2 Count
Iteration 3 Count
Total Count
aerospace
5.77B
261.63M
309.33M
6.34B
9100000
688505
611034
10399539
agronomy
13.08B
947.41M
229.04M
14.26B
15752828
2711790
649404
19114022
artistic… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/FineFineWeb-sample.MapPool
MapPool - Bubbling up an extremely large corpus of maps for AI
MapPool is a dataset of 75 million potential maps and textual captions. It has been derived from CommonPool, a dataset consisting of 12 billion text-image pairs from the Internet. The images have been encoded by a vision transformer and classified into maps and non-maps by a support vector machine. This approach outperforms previous models and yields a validation accuracy of 98.5%. The MapPool dataset may help to train… See the full description on the dataset page: https://huggingface.co/datasets/sraimund/MapPool.MAPBench-V2For more details, please check our project page.
Paper: https://arxiv.org/abs/2601.05432
Repository: https://github.com/AMAP-ML/Thinking-with-Map
cleanvid-15m_map
CleanVid Map (15M) 🎥
TempoFunk Video Generation Project
CleanVid-15M is a large-scale dataset of videos with multiple metadata entries such as:
Textual Descriptions 📃
Recording Equipment 📹
Categories 🔠
Framerate 🎞️
Aspect Ratio 📺
CleanVid aim is to improve the quality of WebVid-10M dataset by adding more data and cleaning the dataset by dewatermarking the videos in it.
This dataset includes only the map with the urls and metadata, with 3,694,510 more entries than… See the full description on the dataset page: https://huggingface.co/datasets/shinonomelab/cleanvid-15m_map.pk-map-statswmap-single-year-maps
WMAP DR5 Single-Year I/Q/U Maps
The preview renders the canonical year-1 K1 TEMPERATURE field in its source
NESTED order, using a Galactic Mollweide projection and a symmetric 99.5th
percentile colour range.
This dataset contains the complete full-resolution single-year I/Q/U release
served by LAMBDA: ten WMAP differencing assemblies for each of nine observing
years. These are year-specific, per-assembly measurements. They are distinct
from wmap-band-maps-9yr, whose five maps… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/wmap-single-year-maps.GSwar3-maps
Warcraft III Community Map Archive
This public dataset preserves community-created Warcraft III maps and campaigns
for interoperability testing, search, research, and long-term access. Files are
deduplicated by SHA-256. Titles and other metadata are extracted with
war3-manager where the format permits.
Search and download individual maps: https://war3-archive.github.io/war3-maps/
Source and issue tracker: https://github.com/war3-archive/war3-maps
Layout… See the full description on the dataset page: https://huggingface.co/datasets/magicwenli/war3-maps.juno-microwave-maps
Juno microwave sky maps and map-space companions
This dataset contains the 2025 LAMBDA release of Juno Microwave Radiometer sky
maps and correlated-noise companions. Each of the 48 FITS bintables is one
Parquet configuration named by its source filename stem. Its column is T,
with the source table shape, float64 or int64 dtype, and row order. The six HDF5 companions add 20 dense-matrix
configurations named by the source filename stem, __, and the HDF5 leaf name.
Each leaf name… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/juno-microwave-maps.planck-frequency-maps-pr4
Planck PR4 NPIPE Frequency Maps
This dataset contains the nine Planck Public Release 4 frequency maps
produced by the NPIPE joint LFI/HFI processing pipeline. It serves the
full-mission, full-channel source tables exactly as released: source
column names, units, fixed-vector shapes, row order, dtypes, and value
bits are preserved. Metadata makes the source representation
self-describing without flattening, renaming, or normalising it.
The included maps are the R4.00 products at… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/planck-frequency-maps-pr4.AetherCode
AetherCode: Evaluating LLMs' Ability to Win In Premier Programming Competitions
Introduction
Competitive programming has emerged as a critical benchmark for evaluating the reasoning and coding capabilities of Large Language Models (LLMs). Despite impressive progress on existing benchmarks, we argue that current evaluations overstate model proficiency, masking a substantial gap between LLMs and elite human programmers. This gap arises… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/AetherCode.planck-frequency-maps-pr3
Planck PR3 Legacy Frequency Maps
This dataset contains the nine Planck Public Release 3 (2018 Legacy)
full-mission, full-channel frequency maps. It serves the source binary
tables exactly as released: source column names, units, scalar shapes,
row order, dtypes, and value bits are preserved. Self-description in
the Parquet metadata resolves the otherwise opaque source identifiers;
it does not rename or normalise them.
The LFI products are the R3.00 maps at 30, 44, and 70 GHz.… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/planck-frequency-maps-pr3.wmap-band-maps-9yr
WMAP Nine-Year Frequency Band Maps
This dataset contains the five WMAP nine-year RAW, rather than
foreground-reduced, coadded frequency-band sky maps. Each source file
contributes a Stokes Maps config containing full-sky Stokes I, Q, and
U measurements and observation counts, and a separate Weight Arrays
config containing its polarisation weights. The nominal frequencies
are K = 23 GHz, Ka = 33 GHz, Q = 41 GHz, V = 61 GHz, and W = 94 GHz.
LAMBDA describes the coaddition: "The… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/wmap-band-maps-9yr.FineFineWeb-test
FineFineWeb: A Comprehensive Study on Fine-Grained Domain Web Corpus
arXiv: Coming Soon
Project Page: Coming Soon
Blog: Coming Soon
Data Statistics
Domain (#tokens/#samples)
Iteration 1 Tokens
Iteration 2 Tokens
Iteration 3 Tokens
Total Tokens
Iteration 1 Count
Iteration 2 Count
Iteration 3 Count
Total Count
aerospace
5.77B
261.63M
309.33M
6.34B
9100000
688505
611034
10399539
agronomy
13.08B
947.41M
229.04M
14.26B
15752828
2711790
649404
19114022
artistic… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/FineFineWeb-test.africa-landuse-11class-v1
MapSpace LSG 2026: Africa 11-Class Land Use
Dataset Details
This dataset contains POI-based land-use estimates for LandScan Global grid cells in Africa. It was generated and is shared by the Geospatial Science and Engineering Division at Oak Ridge National Laboratory.
Curated by: Geospatial Science and Engineering Division, Oak Ridge National Laboratory
Version: v1
License: CC BY 4.0
Geographic coverage: Africa
Spatial grid: 30 arc seconds (1/120 degree)
CRS:… See the full description on the dataset page: https://huggingface.co/datasets/MapSpaceORNL/africa-landuse-11class-v1.asia-landuse-11class-v1
MapSpace LSG 2026: Asia 11-Class Land Use
Dataset Details
This dataset contains POI-based land-use estimates for LandScan Global grid cells in Asia. It was generated and is shared by the Geospatial Science and Engineering Division at Oak Ridge National Laboratory.
Curated by: Geospatial Science and Engineering Division, Oak Ridge National Laboratory
Version: v1
License: CC BY 4.0
Geographic coverage: Asia
Spatial grid: 30 arc seconds (1/120 degree)
CRS: EPSG:4326… See the full description on the dataset page: https://huggingface.co/datasets/MapSpaceORNL/asia-landuse-11class-v1.ovro-lwa-sky-maps
OVRO-LWA Sky Maps
These eight brightness-temperature maps were made by
Tikhonov-regularized m-mode analysis imaging with the Owens Valley Radio
Observatory Long Wavelength Array. Each has 50,331,648 NSIDE-2048
pixels, with coverage nominally north of declination -30 degrees. They
are harmonically filtered products: each map has a00 = 0, all m = 0
spherical-harmonic modes removed, and m = 1 modes with l > 100
filtered to suppress terrestrial interference. Large-scale structure… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/ovro-lwa-sky-maps.FineFineWeb-validation
FineFineWeb: A Comprehensive Study on Fine-Grained Domain Web Corpus
arXiv: Coming Soon
Project Page: Coming Soon
Blog: Coming Soon
Data Statistics
Domain (#tokens/#samples)
Iteration 1 Tokens
Iteration 2 Tokens
Iteration 3 Tokens
Total Tokens
Iteration 1 Count
Iteration 2 Count
Iteration 3 Count
Total Count
aerospace
5.77B
261.63M
309.33M
6.34B
9100000
688505
611034
10399539
agronomy
13.08B
947.41M
229.04M
14.26B
15752828
2711790
649404
19114022
artistic… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/FineFineWeb-validation.dalton-lv-meshes
Dalton et al. (2023) Left-Ventricle Meshes — DOLFINx XDMF
Finite-element meshes of 2967 synthetic left-ventricle (LV) geometries, converted
from the raw node data of
D. Dalton, H. Gao, D. Husmeier, "Emulation of cardiac mechanics using Graph
Neural Networks", Computer Methods in Applied Mechanics and Engineering 401
(2023) 115645. Dataset: Zenodo 7075055
· Code: https://github.com/dodaltuin/passive-lv-gnn-emul
The upstream download contains only a node-adjacency graph plus… See the full description on the dataset page: https://huggingface.co/datasets/mapengfei/dalton-lv-meshes.wikimedia-commons-maps_beirThis is a copy of https://huggingface.co/datasets/jinaai/wikimedia-commons-maps reformatted into the BEIR format. For any further information like license, please refer to the original dataset.
Disclaimer
This dataset may contain publicly available images or text data. All data is provided for research and educational purposes only. If you are the rights holder of any content and have concerns regarding intellectual property or copyright, please contact us at "support-data (at)… See the full description on the dataset page: https://huggingface.co/datasets/jinaai/wikimedia-commons-maps_beir.ADFA_Mappingakari-fir-all-sky-maps
AKARI FIR All-Sky Maps
These four configurations contain the native-resolution AKARI
Far-Infrared Surveyor all-sky intensity maps at 65, 90, 140 and 160
microns. They preserve the NSIDE-4096 HEALPix renderings made by CADE
and frozen by NASA LAMBDA.
Instrument
AKARI Far-Infrared Surveyor
Bands
N60, WIDE-S, WIDE-L and N160
Wavelengths
65, 90, 140 and 160 microns
Pixelisation
HEALPix NSIDE 4096; 201,326,592 pixels
Ordering
NESTED
Coordinate frame
Galactic… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/akari-fir-all-sky-maps.FineLeanCorpus
FineLeanCorpus: A Large-Scale, High-Quality Lean 4 Formalization Dataset
🔍 Overview
FineLeanCorpus is the largest high-quality dataset of natural language mathematical statements paired with their formalizations in Lean 4, comprising 509,356 entries. This dataset is designed to advance research in mathematical autoformalization—the translation of natural language mathematics into formal, machine-verifiable code.
The corpus is distinguished by its:
Scale: 509K entries… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/FineLeanCorpus.iris-infrared-maps
IRIS Infrared Maps
The Improved Reprocessing of the IRAS Survey (IRIS) provides co-added
infrared sky-brightness maps at 12, 25, 60 and 100 microns. These four
configurations are the authors' native-resolution NSIDE-2048 nohole
HEALPix products, frozen by NASA LAMBDA. HCON1, HCON2 and HCON3 were
co-added, and DIRBE data fill the roughly two per cent of the sky not
observed by IRAS.
Bands
12, 25, 60 and 100 microns
Pixelisation
HEALPix NSIDE 2048; 50,331,648… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/iris-infrared-maps.cobe-dmr-four-year-sky-maps
COBE DMR four-year sky maps
Regenerate the preview from the published Parquet data with
python tools/render_hub_preview.py cobe-dmr-four-year-sky-maps datasets/cobe-dmr-four-year-sky-maps/preview.png.
This dataset contains the twelve COBE Differential Microwave Radiometer
four-year Project Data Set sky maps served by LAMBDA. The independent A and B
radiometers at 31.5, 53, and 90 GHz are each provided in the source ecliptic
and Galactic quadrilateralized spherical-cube… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/cobe-dmr-four-year-sky-maps.arcade2-sky-maps
ARCADE 2 Sky Maps
Regenerate the preview from the published Parquet data with
python tools/render_hub_preview.py arcade2-sky-maps datasets/arcade2-sky-maps/preview.png.
These twelve maps are the calibrated sky-map measurement release from the
July 2006 flight of ARCADE 2. The cryogenic differential radiometers compared
the sky with a full-aperture blackbody calibrator. The maps retain absolute sky
intensity after removal of the CMB dipole; no other Galactic or cosmological… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/arcade2-sky-maps.central-america-landuse-11class-v1
MapSpace LSG 2026: Central America 11-Class Land Use
Dataset Details
This dataset contains POI-based land-use estimates for LandScan Global grid cells in Central America. It was generated and is shared by the Geospatial Science and Engineering Division at Oak Ridge National Laboratory.
Curated by: Geospatial Science and Engineering Division, Oak Ridge National Laboratory
Version: v1
License: CC BY 4.0
Geographic coverage: Central America
Spatial grid: 30 arc… See the full description on the dataset page: https://huggingface.co/datasets/MapSpaceORNL/central-america-landuse-11class-v1.qmap-raw-maps
QMAP raw sky maps
The preview projects the source x, y, and z unit vectors onto a north
polar view and colours each source row by T; the dataset itself retains only
the five source columns.
Regenerate it from the published Parquet data with
python tools/render_hub_preview.py qmap-raw-maps datasets/qmap-raw-maps/preview.png.
This dataset contains the six raw QMAP sky-map tables served by LAMBDA. QMAP
observed a region around the North Celestial Pole during two balloon flights
in… See the full description on the dataset page: https://huggingface.co/datasets/astro-legacy-archive/qmap-raw-maps.MAPS-ClinVar-VKS-Embeddings-L80
MAPS ClinVar/VKS ESM-C layer-80 difference fields
Mutant-minus-wild-type difference fields at block 80 of ESM-C 6B for
all 200,913 human missense variants of known clinical
significance in the MAPS ClinVar/VKS set: the complete
12,565-variant held-out test split and the complete
188,348-variant training pool, no sampling on either side.
260.8 GB of raw fp16 payload, 154.0 GB on disk in 262 shards, one parquet row per variant, every row self-describing — no join with
any other… See the full description on the dataset page: https://huggingface.co/datasets/ching-goodfire/MAPS-ClinVar-VKS-Embeddings-L80.