SAVRN Model Hub
AI Training Datasets
Each dataset with its card, its structure (every split, row count and column type), its files, its license, and the models that disclose training on it.
Updated 2026-10-18 · How the library is built
3,164 datasets, sorted by most downloaded.
Independent, daily-updated trust & safety data for the x402 agent-payment economy (HTTP 402 + stablecoins on Base) and the MCP server ecosystem — by PulseFeed. AI agents increasingly pay for APIs autonomously over x402 and connect to MCP servers that can run code on install. But 20% of listed x402 endpoints are dead or invalid, and "live" is not the same claim as "payable": of 24131 endpoints that return a valid 402, only 22426 across 1904 operators are things an agent can actually buy — the rest are documentation placeholders, testnet sandboxes priced in dollars, or payment schemes outside the x402 spec. Both figures are in ecosystem.json as healthyEndpoints and payableEndpoints; compare…
Model evaluation results for System One Mosaic Benchmark (S1MB), combining specialized Choice, Noul, and Score tasks. Each / folder represents one leaderboard entry and contains metadata.json plus per-benchmark.json.xz files. Results retain their evaluation settings and model, dataset, and evaluator provenance. To add or update results, follow the for evaluation, export, validation, and opening a pull request to this dataset. See the S1MB project for code and documentation.
Unmodified copies of public EEG motor imagery datasets in the MOABB cache layout. Files extracted from an upstream archive keep their bytes; Weibo2014's also take the names MOABB gives them. Zhou2016's saved Zenodo record is the one generated file; see its section. Every dataset loads the same way with MOABB 1.7: 1. Download the files you need with snapshotdownload, narrowed by allowpatterns. 2. Point MOABB at them through its MNEDATASETS PATH environment variable. 3. Call moabb.setdownloadprovider("upstream") so MOABB uses the local files instead of contacting NEMAR. Each example below loads subject 1. The pattern for all of a dataset's subjects is given in its text. Each subject's two…
AfriNews is a comprehensive dataset of news articles collected from over 230 prominent news websites across 47 African countries. - Default (default): Contains all articles across all countries and languages. - Country + Language Subsets ( ): Articles from a specific country in a specific language (e.g. CameroonCMfr, CameroonCMen, MoroccoMAar, GhanaGHen). Curated under the AfriSpeech initiative. For research access and feedback, contact: https://huggingface.co/AfriSpeech
Append-only teacher-output cache used to distill a ternary Hindi ASR student. It stores, per audio clip, ONLY: - rowid (a deterministic source-shard/position pointer), - top-k (k=64) CTC logits (vocab id + log-prob per kept entry, blank always kept), - selected encoder hidden states (last-3 blocks: layers 14 / 15 / 16), fp16. No audio and no ground-truth transcripts are stored or redistributed. - Source corpora (audio & transcripts NOT redistributed here; used only to compute teacher outputs): - ai4bharat/Kathbath (hindi) — CC-BY-4.0. - ai4bharat/Shrutilipi (hindi) — CC-BY-4.0. CC-BY-4.0. Redistributing the derived teacher outputs (logits + hidden states + ids) is not redistribution of the…
Co-registered Sentinel-1 / Sentinel-2 pairs for self-supervised pre-training in Earth observation. Generated 2026-08-09. Every entry is a pair: one Sentinel-2 optical tile and one Sentinel-1 radar tile over the same ground, written to the same grid — identical bounding box, identical CRS (the tile's local UTM zone) and 10 m pixels — so the two rasters correspond pixel for pixel with no resampling on your side. The two acquisitions are at most three days apart; the realised gap is recorded per pair. Each tile is 512 × 512 pixels at 10 m, so it covers 5.12 × 5.12 km. The remaining Sentinel bands are not shipped, to keep the corpus trainable on ordinary hardware. They remain recoverable: every…
Chronicling America (US Library of Congress) - Parquet Dataset A high-performance, columnar Apache Parquet dataset containing digitized, OCR-extracted historical American newspapers from the US Library of Congress Chronicling America / National Digital Newspaper Program (NDNP). Produced by streaming and transmuting massive Library of Congress preservation archives (.tar.bz2, METS/MODS, and ALTO XML) into compact, issue-level Parquet shards partitioned by state, newspaper
Surface-normal control videos of annotated indoor scenes, with per-pixel instance ids, camera trajectories, six-section captions, LynnReal-Omni RGB renders, the scene meshes, and the packed MiniMax-H3 Ref2VA training split. Every clip is 832x480, 24 fps, 124 frames (5.17 s). The clips are InsWindowFilt/selectfinal.json from this repository's earlier layout (commit b9305a2, since replaced), copied as selectfinal.json: 10,219 clips over 2,950 scenes that passed 1. black fraction, edge density, instwinmin >= 3 (24,560 of 46,309 renders); 2. trajectory dedup within a scene: position Hausdorff = 5 instances of at least 0.1% area, c0 >= 0.6 of their area seen later. One clip has one key…
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.