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.
11,122Models
3,164Datasets
285Papers
6,792Publishers
11,852Sourced relationships
Updated 2026-10-18 · How the library is built
3,164 datasets, sorted by most downloaded.
761,495 citations from ten search-augmented LLMs answering 11,200 Stack Exchange questions across 28 sites, collected March 26–April 9, 2026. Replace analysismaster with any config below. Use revision= to pin a dataset version. Each config maps to data/.parquet. Join tables using queryid and urlid; citid identifies citations. Field metadata are in croissant.json. Score lookups and site definitions are in scoringmatrices/ and sitetopology/. IPA scores use QI and SP; SS scores use SD and ST; ASF scores use the numeric label suffix. Classification validation contains 1,000 items and 3,000 human labels. Matrix validation contains 900 expert ratings: 30 IPA cells and 60 SS cells, each with ten…
This dataset evaluates design patent Freedom-To-Operate (FTO) image retrieval. Given a query image, a system ranks design patents; targetpns contains the annotated target publication numbers. The public test split contains 80 invalidation-proceeding records selected with ispublic = 1. It contains no e-commerce complaint samples. The 80 query PNGs are embedded in data/test.parquet. The data/image/ files in the repository belong to an earlier release and are not the source for the current test split. querypn is null for 27 samples. Use the query image as the input; do not infer a missing query PN from its filename or target PNs. Use targetpns as the patent-number ground truth. Hit Rate @ K is…
This dataset contains clean, structured public data exported directly from production runs of Apify actors. It serves as a benchmark and sample for lead qualification, market intelligence, research, and machine learning pipelines. Real-time snapshot of top live streaming channels on Twitch. Includes channel display names, game/category names, stream titles, concurrent viewer counts, broadcaster languages, stream start timestamps, and thumbnail URLs. Generated via Apify Actor captainhandsome/twitch-live-streams-scraper. - streamlabel: (e.g. 1 million subscriber celebration - happyhappygal) - channelurl: (e.g. https://www.twitch.tv/happyhappygal) - channelname: (e.g. happyhappygal)…
GR00T-N1.7 LIBERO-X backbone features — 90-task fine-tune (LEVEL1-3) Aligned rollouts of a LIBERO-X fine-tune of GR00T-N1.7 (rohansiva/gr00t-libero-x-90task) on the LIBERO-X simulator, over the exact 90 tasks that checkpoint was fine-tuned on (30 tasks × 3 difficulty levels, LEVEL1–LEVEL3). Train and eval task sets are identical by design, so this is an in-distribution dataset for the checkpoint. 90 tasks × 20 rollouts = 1,800 episodes (600 per level). Every GR00T inference
This dataset contains clean, structured public data exported directly from production runs of Apify actors. It serves as a benchmark and sample for lead generation, labor market intelligence, and compliance verification. Real verified extract of HVAC repair, installation, and commercial contractor businesses across Phoenix, Arizona including business names, phone numbers, full addresses, ratings, review counts, Google Maps URLs, and website domains. Generated via Apify Actor captainhandsome/google-maps-business-search. - name: (e.g. Ken Muncy Air Conditioning) - placeurl: (e.g. https://www.google.com/maps/place/Ken+Muncy+Air+Conditioning) - placeid: (e.g.…
Raw per-sample results from MOevals runs. One directory per upload; nothing here is aggregated — the Parquet and the.eval logs are the primary evidence, the scorecard is a summary of them. This repo was private when this card was written. Results repos are public unless dangerous or unpublished-sensitive (docs/COMMANDS.md); the flip is the owner's call, made on the Hub. scripts/runall.sh uploads to runs/ /; a direct uploadresults.py call without --path-in-repo uploads to - /. Uploaded by scripts/uploadresults.py in the MOevals repo, which is the source of truth for how these were produced. Judge caveats (calibration status, fake-mode bans) live in the run's config and scorecard, not here.…
VEX-Bench scores the verification burden of LLM-generated misinformation after elicitation, not jailbreak success alone: - D1 Checkability - D2 Harm Significance - D3 Source Credibility Signals - D4 Imposter Legitimacy (field: presentation) - D5 Verification Cost 7 models × 7 methods × 2 tasks × 6 domains × 10 topics = 5,880 cells. No train/test split: every config uses split full. This snapshot is sanitized and does not include rewrite source-article bodies. Models: anthropic/claude-sonnet-4, deepseek/deepseek-v4-pro, google/gemini-3.1-pro-preview, moonshotai/kimi-k2.5, openai/gpt-5.4, qwen/qwen3.5-flash-02-23, x-ai/grok-4.1-fast. Methods: direct, isc, misinfoqa, poisonedrag, disinfocap…
A dated, public record of which businesses AI engines name when people ask for a recommendation. Each row in the answers table is one question on one date to one engine. Every claim owes the future a receipt. Edition 1 is final. Data through 2026-10-13. Counts: 73 questions, each asked in the same words on every run day it was in the roster (55 market or category questions and 18 questions asked to see whether one named business comes up); 25 US local markets plus nationwide questions (27 market labels as written in the roster); 37 niche labels as written in the roster; API engines asked: anthropic, gemini, openai, perplexity, plus Google results; 42 run days from 2026-09-01 to 2026-10-13.…
Results accompanying Evaluating Tool-Augmented PII Detection in Small Language Models, a technical report (not peer reviewed). The current files replace ambiguous historical score fields with an explicit report scoring contract. No model inference was rerun. All 66,000 prediction records remain; labels, offsets and record identities are preserved from the prior Hub release. Detector prediction order is preserved as published. Entity text is redacted, including text previously present in the detector split. Prior repository revisions remain historical and are superseded for evaluation. The standalone detector previously lacked fallback gold for 868 examples in local scoring. Correct gold is…
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.