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.
A documented Charts data-preparation workflow for Image Depth 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 testing…
Three archives from Meta's Seamless Interaction Naturalistic training split, used in our CMU 11-777 project proposal. The subset contains 37 participant-side recordings from 27 participant identifiers, totaling 3.20 GB. Each recording includes MP4, WAV, NPZ and JSON files. Archives were selected in order; this is a convenience sample. - naturalistic/train/0012/0000.tar - naturalistic/train/0012/0001.tar - naturalistic/train/0012/0002.tar - manifest.json: file sizes, SHA256 hashes and source URLs. The archives are unchanged copies of the source files. Recording counts are before motion-quality and synchronized-window filtering. Seamless Interaction is provided by Meta under CC BY-NC 4.0.…
mockbuster 识别项目的两侧限定样本（此包供云端 GPU 计算标题 / 简介 / 海报相似度）。 双侧有海报且有 overview，共 70,733,031 对。两侧同时施加样本定义后： - 分片顺序即文件名字典序（part00 → part07）。 - 整包校验：sha256sum postersrestricted.tar 应等于 15d588589553f7edee12c2ca9580adceebb7d62a02640897a3523008b3500ee1。 - 逐片校验见 partsmanifest.tsv；tar 内结构为 posters/.jpg（原创）与 posterscand/cand.jpg（候选）。 - 标题相似度：用北美英语发行片名（origtitle / candtitle）。 - 简介相似度：用 origoverview / candoverview。 - 海报相似度：posters/.jpg 与 posterscand/cand.jpg。 - pairs.tsv.gz 的 daysdiff / arrival 基于 usanymin（含影展首映）；商业发行日口径见同表 canduscommercialmin 列。 - 候选侧池为英语池（originallanguage=='en'），不含非英语候选。
Dataset phục vụ cho nghiên cứu và huấn luyện mô hình AI. Dataset được lưu dưới dạng CSV.
MSMARCO docs + queries with pre-computed 768-dim label embeddings. - docs were kept only if they tokenize to ≤ 128 tokens (bert-base-uncased vocab).
39,996 synthetic doctor/patient dialogues, generated from for supervised fine-tuning. Each one turns a single question-and-answer row of the source corpus into a conversation where a clinician has to elicit the case before answering it. The cases are included so the run can be reproduced, re-scored, or regenerated with another model. A dialogue file on its own cannot be. Three participants, isolated by construction rather than by instruction: - doctor — sees the opening line and the question it must answer. Nothing else. It chooses each turn whether to ask, to examine, or to conclude, and deciding it has enough is the capability being taught. - patient — holds the spoken facts and discloses…
This dataset contains text-and-audio pairs selected for TTS training. The initial audio/uzbek split has 2,184 utterances. Other languages and additional rows can be added as separate, append-only shards. Each row retains the complete structured text specification alongside audio and the OmniVoice fields utt, text, spk, and instruct. ttstaggedtext preserves the local delivery and vocal-event controls separately from the clean text field. For this release, the saved synthesis request matches the selected transcript and speaker, and each WAV passed structural checks. The audio has not been independently transcribed or human-rated; qastatus records that distinction. The matching source…
Preparation notes and schema examples for Ocean tasks using Multimodal3 data. Full source material is intentionally not bundled, so provenance and licensing remain explicit. - builddataset.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 testing the…
This data card accompanies a lightweight Movie Posters loader for Multimodal3 metadata. It is meant for pipeline inspection, source adaptation, and reproducible split preparation. - 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 testing the…
This dataset card aims to be a base template for new datasets. It has been generated using this raw template. Users should be made aware of the risks, biases and limitations of the dataset.
Prepared by Peter Kirby for the Multilingual Universal Style Embeddings (MUSE) project. These are supplemental authorship training sources with account-grouped train/validation splits. Each row preserves one supplied source text after recorded outer trimming. No instruction prefixes, model outputs, or contrastive pairs have been inserted into the text. - macauthorship: MAC full v1, with 59 verified wiki language labels. - exordeauthorship: the complete selected Exorde view. - exordeauthorship: 77 Exorde language/platform views across 41 languages. For example, exordeauthorshipenx.com. The aggregate Exorde view references the same shards as its individual configurations. Count physical rows…
Dataset · Tabular classification
nairobi-longitudinal-healthcare-utilization
Synthetic population data for predicting healthcare demand across time. A genuine balanced longitudinal panel: the same 10,000 synthetic Nairobi residents, tracked at 3 annual observation dates (2023-09-01 / 2024-09-01 / 2025-09-01 = Train / Validation / Test) — 30,000 person-period rows, 38 columns. Each row carries up to 24 months of prior health, insurance, household and employment history as of that date, plus what happened to that same person over the following 12 months. This links the same synthetic person across health, insurance, household, education and employment domains, over time — not a single disease table. Marketplace question: Who is most likely to use healthcare — or…
4,000 distinct controlled game states with five questions each (20,000 questions). The generator supplies the complete remaining rank counts and computes outcome distributions by exact branch enumeration using float64 dynamic programming. These are simulated game states, not recorded casino games. - Draw without replacement from a subset of one standard 52-card deck. Ranks are A,2,...,9,10-value; 10/J/Q/K are combined. - The state gives player cards, total, usable ace, dealer upcard and all remaining drawable counts. Visible cards have already been removed. - There is no predealt dealer hole card and no peek. The dealer draws after the player stops. Player bust loses immediately. - Only hit…
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.