15,089 real market decision times, with three direction questions per time (45,267 correlated questions): 1, 5 and 15 minutes. No synthetic market rows or generated prices are used. Derived from Martin Søgaard Nielsen (martinsn), High Frequency Crypto Limit Order Book Data, version 1, BTC1min.csv. The publisher reports Coinbase websocket collection and declares CC0: Public Domain. The source contains 17,113 rows from 2021-04-07 11:33:41 UTC to 2021-04-19 09:54:00 UTC (13 UTC dates, about 12 elapsed days). No ETH or additional dates are included. The unchanged source ZIP and provenance are included under trading/. Original source and derived dataset records are distributed under CC0-1.0…
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.
Released by AI Lab Pvt. Ltd., Nepal. Text only: no audio is included. For every clip of our Nepali–English code-mixed test set (4,610 clips, 9.1 h) and of the FLEURS Nepali test set (586 digit-free clips), this dataset gives the reference transcript and the output of 7 speech recognition systems: SraVaani-1.0 zero-shot and fine-tuned for 6 and 12 epochs, and Whisper large-v3-turbo and IndicConformer, zero-shot and fine-tuned. Use it to inspect errors, compare models clip by clip, or recompute any metric. WER is computed over the whole set (total word errors divided by total reference words) on the normalised text. FLEURS writes English loanwords in Devanagari while our models write them in…
Native full-latent GT fitting Incremental private results for all 8627 train and 473 validation audited generated-MPM clips. Posterior mean initialization; frozen NeuROK150k; 128x1024 free latent parameters; Adam lr0.1,200 fixed updates per frame,final iterate. No GeoDRAS basis or rank constraint. Rank20 and other representations are downstream experiments. Partial until all9100 completed reports are present. Every clip contains checksums,split provenance and full latent arrays.
This dataset provides comprehensive, multi-season panel data for the National Hockey League (NHL), spanning over 17,000 games, 600,000 player-game appearances, and millions of play-by-play event coordinates from 2013 to the present day. Unlike fragmented yearly CSV dumps, this repository organizes NHL data by domain entity and data type into high-performance Apache Parquet tables with standardized schemas, UTC timestamps, and validated primary keys. Automated daily GitHub Actions workflows incrementally append and update incoming game records from the NHL Edge API. Data is scraped and transformed from the official NHL Edge API and market data providers. Maintained via the NHL-Analytics…
The 94-stack: an expanding base-93 symbol system (Gödel-numbered data-as-code) grown from a design conversation, building toward a deterministic TTRPG engine. Every file is standalone Python 3, stdlib only, and ships with its own verification suite (exit 0 = all checks pass). Design doc: see the conversation artifacts ("The Expanding Base94 Symbol System") for the full specification — encoding, zones, chaos dial, epistemics, agents, spins, hashing, the Watcher, the palette, the fly circuits, and functors.
A documented NLP Summarization data-preparation workflow for Audio Text records. The bundled rows demonstrate the schema and validation path rather than pretending to be a full training corpus. - loaddata.py — loading, cleaning, and split preparation code. - datasetinfos.json — schema and split metadata. - metadatasample.jsonl — small, human-readable records for checking the schema. - README.md — data card and usage notes. Before using the prepared data, verify source licenses, duplicates across splits, missing values, label balance, and modality-specific corruption. Record the source version and every filtering rule so a later run can reproduce the same rows. The repository is suitable for…
This dataset contains Norwegian speech data from NRK TV broadcasts (norge-rundt, supernytt), processed for automatic speech recognition (ASR) evaluation and research. - id: Unique chunk identifier - audio: Audio data (WAV format, 16kHz) - text: Original transcription text - textnormalized: Normalized text (lowercase, standardized) - durationseconds: Audio duration - chunktype: Type of chunk (subtitlealigned or fixedduration) - episodeid: Source episode identifier - programname: NRK program name - starttime: Start time in source video (seconds) - endtime: End time in source video (seconds) TBD - Please verify NRK terms of use before distribution. If you use this dataset, please cite: For…
Not a conventional browser. OFFDIGITZ is a chrome-less, unrestricted GeckoView browser whose entire interface is a command palette — with a built-in AI assistant. No URL bar. No toolbar row. No tab strip. Content is edge-to-edge, full screen. One floating ⌘ button summons the command palette — everything routes through it: - Type anything → Go to … / Search DuckDuckGo / Ask AI A thin loading bar hugs the top edge — that's the only persistent indicator. Engine-level limit (not embedder-fixable): Gecko validates add-on signatures at install time — unsigned.xpi installs are refused by the engine regardless of app settings. Privileged extensions ship inside the APK (OFFDIGITZ's dev console does…
This dataset contains 100 original procedurally modeled abstract objects, each with a unique nonce name and a fixed, entirely fictional maker record. It provides 10,000 RGB images at 512×512, one portable GLB mesh and one preview per object, and paired data for visual understanding and text-to-image generation. The objects and all maker records are newly constructed for this dataset; no CLEAR identities, biographies, images, or external meshes are included. Each object has 60 training views, 20 held-out camera views, and 20 held-out appearance views with changed background and lighting. A held-out appearance view may share a camera direction with training. The same geometric object appears…
Evaluation scores for the two OpenEuroLLM 32B cooldown (annealing) runs, every 2000 iterations from 100k to 120k, scored on the engbasemain and deubasemain suites of base-eval via Weights live in openeurollm/oellm-32b, one branch per checkpoint ( iter ). Both runs share globalbatchsize 4096 × seqlength 4096 = 16,777,216 tokens/step, so tokenstrained = 16777216 × step (120000 → 2.013T). The two runs are on an identical step grid, so they can be compared point by point. Same schema as openeurollm/prelude-base-eval-scores. - model, step, tokenstrained, modeldisplayname, modelpath - task, metric, score, scorestderr, numfewshot, language, nsamples - tasktype — benchmark (leaf task), taskgroup…
A structured, bilingual (EN/AR) catalog of official statistical publications from Oman's National Centre for Statistics and Information (NCSI), harvested from - 1,844 publications across 15 subject areas, each with bilingual titles, direct EN/AR PDF links (and XLS where available), category, and publication dates. - Every file link was verified (HTTP status + size) at harvest time: 3,703/3,706 files confirmed reachable (~25.0 GB of official reports). - other - v0.2+: data.gov.om indicators, respecting its robots.txt (its SDMX API is disallowed). Published by NCSI, Sultanate of Oman. Redistribution follows the NCSI open-data terms of use (see license link). The catalog itself (metadata…
A bilingual (Arabic + English) plaintext corpus of Wikipedia articles about Oman, built from the official MediaWiki APIs of ar.wikipedia.org and en.wikipedia.org. 1. Walked the Oman root category tree (depth ≤ 2), pruning maintenance and stub categories (Category:Oman on enwiki; تصنيف:سلطنة عمان on arwiki). 2. Ranked candidate articles by page length and kept the top ones. 3. Fetched clean plaintext extracts via the official prop=extracts API (one article per request, globally rate-limited to ~100 requests/minute). 4. Dropped articles with fewer than 200 characters of text. Wikipedia text is CC BY-SA 4.0. This corpus is a derivative collection: any public use must credit Wikipedia…
Dataset chứa 6 giọng tiếng Việt và công cụ speak.py để chạy trên Google Colab với OmniVoice. Mỗi giọng gồm profile.json, voice.pt (prompt cache), audio mẫu và reftext.txt. 1. Lồng tiếng SRT: mở colab/OmivoiceVIColab.ipynb 2. Clone giọng mới: mở colab/OmivoiceVICloneColab.ipynb 3. Đặt HFREPO = " " 4. Runtime → GPU → chạy từng cell - native (mặc định, khuyến nghị): native ≤cap → dùng gap tới cue sau → stretch ≤x1.1 → chồng nhẹ đuôi (crossfade). Không đẩy timestamp SRT — mỗi cue neo đúng start gốc. Rút gọn câu: tự sửa trên file SRT. - cascade: giữ tốc độ tự nhiên, cue dài đẩy cue sau (lệch timeline) - fit: kéo nén tín hiệu vừa khung SRT - strict: ghép theo timestamp (có thể cắt audio)…
This dataset was created using LeRobot.
Source data for the radiograph-dicom-ingest Verifiers environment. Each row is one radiograph with: - the image file exactly as the source distributes it (GRAZPEDWRI-DX 16-bit PNG, FracAtlas JPEG; bytes are not re-encoded), - the source annotation files for that image, verbatim (Pascal VOC XML, YOLO txt, Supervisely JSON, COCO entries), - normalized bounding boxes derived from the source YOLO labels. The dataset contains no DICOM files. It is a research and benchmarking resource and is not for clinical use. Both sources are published under CC BY 4.0, and this subset is released under the same license. Changes from the sources: a subset of images was selected, the files were repackaged into…
Full text of 125,153 machine-learning papers extracted from their PDFs: conference and journal submissions from OpenReview (accepted, rejected and withdrawn where public), used by Open Text Detector. - clean (default): the text layer after cleanup (positioned-clean-v2: hyphenation, headers/footers, page numbers and similar layout artefacts removed). - direct: the raw positioned text layer from poppler (pdftotext -bbox-layout), with page-level tesseract OCR only for pages without a usable text layer. method says which. - omissions: the clean text with non-prose content replaced by markers: figure text, tables, display equations, garbled inline math, number runs and the reference list.…
The v0.3.0 response-policy checkpoint improves authored policy decisions from 127/175 to 147/175 across 35 held-out source programs. At its calibration-selected alert threshold, it flags 14/140 permitted decisions, a 10% false-alert rate on this panel. 26/28 registered checks pass. The DNS diagnostic remains 21/32, with one repaired answer and one newly incorrect answer. These are bounded research results, not field validation or proof of human-versus-agent classification. The current 40-entry collection inventory and catalog preserve all previous entries. The new response-policy configuration contains 139 training, 28 calibration, and 35 development programs with explicit authored…
configname: main datafiles: - main.jsonl.zst - configname: trainable datafiles: - train.jsonl.zst - test.jsonl.zst
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.