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.
This dataset was created using LeRobot.
This sample of Pantheon DiverseUMI contains 25 hours of bimanual tabletop manipulation recorded with two handheld UMI grippers and an exocentric top camera. Each episode is a single task, such as stacking two blocks or passing a ball between grippers. Every episode includes three synchronized videos (top camera and both wrist cameras), per-frame end-effector actions for both wrists, and dense annotations of what the operator does. The sample is split into two halves of roughly equal size: something over, and then recover. Tasks range from 1 second to about 10 minutes, with a median of 7 seconds. Of the 4,588 tasks, 4,391 succeed, 154 fail and 43 are partially completed. For access to more…
GSCBench provides exact graph edit distance (GED) supervision for studying in-collection, cross-collection, zero-shot and few-shot graph similarity models. The GSCBench benchmark data comprise 11 collections, 5,523 graphs and 1,903,452 exact graph pairs. The repository additionally includes a version of PyG LINUX with isomorphic graphs removed (89 graphs and 3,916 pairs), used as auxiliary data for model pretraining. If you use GSCBench in your research, please cite our paper: The tu prefix identifies TU collections, and ogbogbg-code2 identifies the OGB code2 collection. This PyG LINUX subset has isomorphic graphs removed and is provided to reproduce auxiliary model pretraining.…
The Clinical Trials Version History dataset by Anamnesis Data: structured medical and scientific data for biotech, pharma and healthcare research. Public trial registries show only a trial's latest state. This dataset keeps every version, so you can see what a trial said on any past date, or when a sponsor moved a completion date. 604,582 trials (4,444,543 versions), one folder of Parquet files per table (79 tables, 526,178,527 rows). Access is gated; request it above. A free complete trials has the same tables and columns. in the sample holds ready-to-run DuckDB queries, from a trial's current record to serious adverse event rates. They run unchanged against this dataset's which creates…
Reproduction package for the framework paper: frozen-gate machinery, the deployment/long-horizon/domain-shift evaluations, the agentic tiers, and the shared v3 validation campaign. See INDEX.md for the claim→artifact map, MANIFEST.sha256 for per-file hashes, and v3-campaign/EVIDENCE-RELEASE.md for the verification recipe. PENDING.md lists what is not yet packaged and why. Code is Apache-2.0; third-party corpora and model assets retain their upstream licences. This is a partial evidence release; consult PENDING.md before using it. MANIFEST.sha256 covers this snapshot. External evidence gaps remain listed in PENDING.md. See REPRODUCE.md for download and verification instructions. P6 audit…
Uni-VLaT contains 289 demonstrations and 379,397 retained frames from a Unitree G1 without physical hands. This private staging release is being uploaded for Open-X-Tactile. The OXT archives are incomplete until oxtv120261011/UPLOADMANIFEST.json is present. Upload completion does not mean OXT acceptance or completion of semantic and coordinate review. The existing root dataset.zip is the earlier release and is retained unchanged. The new OXT Zarr data are under oxtv120261011/; do not confuse the two formats. Each task folder contains independently extractable ZIPs, one per episode, and its supplementary joint and tactile notes. Each ZIP contains one zarr/.zarr store and a SHA256SUMS…
The custom human-supervised typed-judgment corpus used for OpenJudgement-4B-Preview, developed by Kitani. It adapts documented 2025–2026 source releases into Noul, Choice, and Score decisions. No Mix-v3 rows or local synthetic generators were reused in this corpus. This is an experimental training dataset, not evidence of Jev parity. Coverage is incomplete, source labels can be disputed, and the model remains unfinished. The source-specific terms below continue to apply. The selected model checkpoint was trained for 700 optimizer steps, seeing 44,741 decisions and 49,046,168 input tokens. Those are actual training exposures; the full corpus and split sizes below describe the available data…
PAPE PAPE (Plausible Anomalous Paired Edits) is a paired image-editing dataset for localizing semantic edits. Each sample edits one object in a real photograph twice: a semantically ordinary replacement, and a replacement that introduces one controlled semantic anomaly. Both edits have a binary change mask. Only pairs whose normal and abnormal masks have IoU of at least 0.5 are included. Source photographs come from COCO, ImageNet, and Places. Those photographs stay under the
Caret-based full-line completion samples extracted from permissively licensed Go repositories with the goflc builder (Go go/parser + go/types). Each sample is an exact editor position: leftcontext ends at the caret, targettext is the rest of the physical line (no newline, trailing whitespace excluded), rightcontext follows it. Original source is reconstructable from offsets (byte offsets as used by Go tooling, plus UTF-16 offsets for IDEs); every row carries repositoryid, revision, relativepath and license for attribution. Splits are assigned per repository group (forks/near-copies together); cross-repository duplicate files are kept once. Generated code (// Code generated... DO NOT…
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.