The data construction workflow can be summarized as follows: 1. Deduplicate: The FineWeb dataset is deduplicated using exact deduplication and MinHash techniques to remove redundant data. 2. URL Labeling: Root URLs from FineWeb are counted, and the top 1 million URLs are labeled using GPT-4. This step generates DoI (Domain-of-Interest) Coarse-Grained URLs and DoNI (Domain-of-Non-Interest) Coarse-Grained URLs as seed data sources. 3. Coarse Recall: a. Based on the labeled root URLs, data is sampled for each domain. b. The sampled data is labeled using Qwen2-7B-Instruct, producing 500K DoI Positive Data and 500K DoI Negative Data (note that for N>1 iterations, each 500K samples are composed…
Publicly accessible
apache-2.0
n>1T
GLUE, the General Language Understanding Evaluation benchmark (https://gluebenchmark.com/) is a collection of resources for training, evaluating, and analyzing natural language understanding systems. The leaderboard for the GLUE benchmark can be found at this address. It comprises the following tasks: A manually-curated evaluation dataset for fine-grained analysis of system performance on a broad range of linguistic phenomena. This dataset evaluates sentence understanding through Natural Language Inference (NLI) problems. Use a model trained on MulitNLI to produce predictions for this dataset. The Corpus of Linguistic Acceptability consists of English acceptability judgments drawn from…
Publicly accessible
other
10K<n<100K
SuperGLUE (https://super.gluebenchmark.com/) is a new benchmark styled after GLUE with a new set of more difficult language understanding tasks, improved resources, and a new public leaderboard. An example of 'test' looks as follows. An example of 'test' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. The data fields are the same among all splits. - sentence1: a string feature. - sentence2: a string feature. - idx: a int32 feature. - label: a classification label, with possible values including entailment (0), notentailment (1). - premise: a string feature. - hypothesis: a string feature. - idx: a…
Publicly accessible
other
10K<n<100K
This dataset contains 108,463 human-labeled and 656k noisily labeled pairs that feature the importance of modeling structure, context, and word order information for the problem of paraphrase identification. The dataset has two subsets, one based on Wikipedia and the other one based on the Quora Question Pairs (QQP) dataset. For further details, see the accompanying paper: PAWS: Paraphrase Adversaries from Word Scrambling (https://arxiv.org/abs/1904.01130) PAWS-QQP is not available due to license of QQP. It must be reconstructed by downloading the original data and then running our scripts to produce the data and attach the labels. The text in the dataset is in English. Below are two…
Publicly accessible
other
100K<n<1M
Dataset · Text classification
Google
BoolQ is a question answering dataset for yes/no questions containing 15942 examples. These questions are naturally occurring ---they are generated in unprompted and unconstrained settings. Each example is a triplet of (question, passage, answer), with the title of the page as optional additional context. The text-pair classification setup is similar to existing natural language inference tasks. An example of 'validation' looks as follows. The data fields are the same among all splits. - question: a string feature. - answer: a bool feature. - passage: a string feature. BoolQ is released under the Creative Commons Share-Alike 3.0 license. Thanks to @lewtun, @lhoestq, @thomwolf…
Publicly accessible
cc-by-sa-3.0
10K<n<100K
Large Movie Review Dataset. This is a dataset for binary sentiment classification containing substantially more data than previous benchmark datasets. We provide a set of 25,000 highly polar movie reviews for training, and 25,000 for testing. There is additional unlabeled data for use as well. An example of 'train' looks as follows. The data fields are the same among all splits. - text: a string feature. - label: a classification label, with possible values including neg (0), pos (1). Thanks to @ghazi-f, @patrickvonplaten, @lhoestq, @thomwolf for adding this dataset.
Publicly accessible
other
10K<n<100K
The Adversarial Natural Language Inference (ANLI) is a new large-scale NLI benchmark dataset, The dataset is collected via an iterative, adversarial human-and-model-in-the-loop procedure. ANLI is much more difficult than its predecessors including SNLI and MNLI. It contains three rounds. Each round has train/dev/test splits. English An example of 'trainr2' looks as follows. The data fields are the same among all splits. - uid: a string feature. - premise: a string feature. - hypothesis: a string feature. - label: a classification label, with possible values including entailment (0), neutral (1), contradiction (2). - reason: a string feature. Thanks to @thomwolf, @easonnie, @lhoestq…
Publicly accessible
cc-by-nc-4.0
100K<n<1M
Dataset · Text classification
Ceval
C-Eval is a comprehensive Chinese evaluation suite for foundation models. It consists of 13948 multi-choice questions spanning 52 diverse disciplines and four difficulty levels. Please visit our website and GitHub or check our paper for more details. Each subject consists of three splits: dev, val, and test. The dev set per subject consists of five exemplars with explanations for few-shot evaluation. The val set is intended to be used for hyperparameter tuning. And the test set is for model evaluation. More details on loading and using the data are at our github page. Please cite our paper if you use our dataset.
Publicly accessible
cc-by-nc-sa-4.0
10K<n<100K
BLiMP is a challenge set for evaluating what language models (LMs) know about major grammatical phenomena in English. BLiMP consists of 67 sub-datasets, each containing 1000 minimal pairs isolating specific contrasts in syntax, morphology, or semantics. The data is automatically generated according to expert-crafted grammars. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. An example of 'train' looks as follows. The data fields are the same among all splits. - sentencegood: a string feature. - sentencebad: a string feature. - field: a string feature. - linguisticsterm: a string…
Publicly accessible
cc-by-4.0
10K<n<100K
Dataset · Text classification
Ai2
Given a partial description like "she opened the hood of the car," humans can reason about the situation and anticipate what might come next ("then, she examined the engine"). SWAG (Situations With Adversarial Generations) is a large-scale dataset for this task of grounded commonsense inference, unifying natural language inference and physically grounded reasoning. The dataset consists of 113k multiple choice questions about grounded situations (73k training, 20k validation, 20k test). Each question is a video caption from LSMDC or ActivityNet Captions, with four answer choices about what might happen next in the scene. The correct answer is the (real) video caption for the next event in…
Publicly accessible
unknown
100K<n<1M
The SNLI corpus (version 1.0) is a collection of 570k human-written English sentence pairs manually labeled for balanced classification with the labels entailment, contradiction, and neutral, supporting the task of natural language inference (NLI), also known as recognizing textual entailment (RTE). Natural Language Inference (NLI), also known as Recognizing Textual Entailment (RTE), is the task of determining the inference relation between two (short, ordered) texts: entailment, contradiction, or neutral (MacCartney and Manning 2008). See the corpus webpage for a list of published results. The language in the dataset is English as spoken by users of the website Flickr and as spoken by…
Publicly accessible
cc-by-sa-4.0
100K<n<1M
これに伴い、2009年11月から運用されてきた旧システムは提供終了となり（事実上のサービス終了）、torne や BRAVIA などの家電への対応が軒並み終了する中、当時の生の声が詰まった約11年分の過去ログも同時に失われることとなってしまいました。 そこで 5ch の DTV 板の住民が中心となり、旧ニコニコ実況が終了するまでに11年分の全チャンネルの過去ログをアーカイブする計画が立ち上がりました。紆余曲折あり Nekopanda 氏が約11年分のラジオや BS も含めた全チャンネルの過去ログを完璧に取得してくださったおかげで、11年分の過去ログが電子の海に消えていく事態は回避できました。 しかし、旧 API が廃止されてしまったため過去ログを API 経由で取得することができなくなり、またアーカイブされた過去ログから見たい範囲のログを探す場合も、アーカイブのサイズが合計約 150GB もあることから、とても以前のように手軽に過去ログに触れることはできなくなってしまいました。 このデータセットでは、ニコニコ実況のすべての過去ログを後世に残すべく、Nekopanda 氏が配布されていた旧ニコニコ実況の 2020/12/15 までのすべての過去ログに加え、コミュニティでの実況番組も含めた新ニコニコ実況、さらに 2024/06/10 からは実況用代替コメントサーバーである NX-Jikkyo の当日分の過去ログを5分に1回収集し、随時反映しています。
Publicly accessible
mit
A continuously updated, de-identified corpus of Ukrainian Telegram news and the discussion around it, published by the Ukrainian non-profit Aisberg (ГО «АЙЗБЕРГ»). It comes in two layers. The first is the raw monthly stream: every post from a fixed set of public news channels, with its reactions and its comment thread. The second is the analysis behind every report Aisberg publishes: posts from different channels grouped into one event, the manipulation techniques found in the coverage, who published first and by how many minutes, view and reaction curves over the first hour, comment sentiment, and a fact-check with a ten-label verdict and the full evidence chain behind it. Free to use…
Publicly accessible
cc-by-4.0
100K<n<1M
Real-time feed of phishing, crypto drainer, and scam domains detected by PhishDestroy. Updated hourly from GitHub. MIT License - Free for commercial and non-commercial use.
Publicly accessible
mit
100K<n<1M
Real human-labeled data, reformatted into System One questions — with human label distributions wherever they exist. 22 configs · 166,054 rows · 22,773 on the same 22,773 test records. Down and to the right is better; the stars are open models fine-tuned on their own decision readout with a coherence penalty. Contents — model (TypeSafe's Jev, or any open model Jevified by the engine) does not write text. It reads a state, answers typed questions, and returns probability distributions your code can branch on. The product claim is calibration: an answer given 0.8 should be right about 80% of the time. Most benchmarks can only check the argmax; jev-bench checks the distribution — four configs…
Publicly accessible
other
100K<n<1M
Continuous knowledge datasets from the IDA Dataset Factory. - {name}.csv · {name}.jsonl · {name}.parquet (when available) · README.md See repository LICENSE. Public knowledge from trusted open sources.
Publicly accessible
other
1K<n<10K
G
Dataset · Text classification
Gowda
CodeXGLUE Clone-detection-BigCloneBench dataset, available at https://github.com/microsoft/CodeXGLUE/tree/main/Code-Code/Clone-detection-BigCloneBench Given two codes as the input, the task is to do binary classification (0/1), where 1 stands for semantic equivalence and 0 for others. Models are evaluated by F1 score. The dataset we use is BigCloneBench and filtered following the paper Detecting Code Clones with Graph Neural Network and Flow-Augmented Abstract Syntax Tree. - semantic-similarity-classification: The dataset can be used to train a model for classifying if two given java methods are cloens of each other. - Java programming language An example of 'test' looks as follows. In the…
Publicly accessible
c-uda
1M<n<10M
A continuously-growing collection of AI text detection predictions with optional user feedback, generated from the AI Text Detector Space. Every time someone analyzes text or a URL on the Space, the prediction is appended to this dataset. Users can also click "Correct" or "Incorrect" to provide feedback, which gets stored alongside the prediction. Or with the datasets library: 1. Model evaluation — see how the detector performs on real-world text submitted by users 2. Retraining data — records with feedback provide ground-truth labels for continuous learning 3. Failure analysis — examine where the model is wrong (feedback == "incorrect") 4. Distribution analysis — what kinds of text/URLs…
Publicly accessible
apache-2.0
n<1K
R
Dataset · Text classification
Rr
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…
Publicly accessible
other
Corpus completo derivado de TucanoBR/GigaVerbo, classificado com educability-norberto-mini-4class-v1. As colunas originais text, label, probs e metadata são preservadas. educabilitystatus=model identifica uma previsão do classificador. Textos nulos, vazios, somente espaços ou sem tokens recebem classe 1 e vetor [1, 0, 0, 0] por regra, com educabilitystatus=ruleemptytext; esse vetor representa uma decisão determinística, não confiança do modelo. educabilityreason identifica o motivo. Falhas isoladas ficam como pendingerror, com classe e probabilidades nulas, até serem resolvidas. educabilityclass e educabilityprobabilities registram a classe (1–4) e as quatro probabilidades da anotação…
Publicly accessible
other
O RepoJus é um corpus consolidado de decisões judiciais e administrativas brasileiras obtidas de fontes públicas oficiais. O snapshot v2 reúne 4.106.391 registros canônicos, sem duplicatas de normalizedhash, em 4.666 shards Parquet determinísticos (39.290.819.138 bytes). O corpus foi estruturado para pesquisa jurídica, recuperação de informação, RAG, classificação, sumarização, extração de entidades e treinamento ou avaliação
Publicly accessible
cc-by-4.0
1M<n<10M
Dataset · Text classification
Nyaamoe
NICOLOG アニメコメントアーカイブ commeonやNCOverlayでコメント付きのアニメを楽しもう！
Publicly accessible
mit
Dataset · Text classification
Vinkius
Welcome to the Vinkius Open Data Initiative. We are opening access to the Vinkius connector catalog. This repository provides automatically updated documentation for 10,672 unique connectors for AI agents. This highly structured corpus is designed specifically for AI researchers, data scientists, and language model developers. It provides a robust foundation for advancing artificial intelligence research in the following domains: Extracted programmatically from the Vinkius registry, each record maintains strict structural integrity. The dataset is rigorously indexed with categorical taxonomies, precise tool support configurations, and environmental metadata to ensure a noise-free corpus.…
Publicly accessible
cc0-1.0
1K<n<10K
A land-use / land-cover relevance label for every sentence of NoeFlandre/osm-polygon-wikidata-and-wikipedia (revision d6fc058). The input is mirrored here unchanged; labels and model outputs are separate tables that join back to it. In progress: 100 of 761 input files fully labelled, 661 more partially, 74,791,929 rows, 165 unique texts have their full model outputs published so far (outputs ship with the first complete file that contains the text; labels for the other sentences are in labels/). 928,088 rows (1.2% of all rows) have no Each hexagon is an H3 cell (resolution 3) holding the yes/no sentences whose document's first linked polygon (smallest polygonid with lat/lon) falls in it.…
Publicly accessible
odbl
A structured, machine-readable database of 1330 confirmed AI agent security incidents, collected and classified automatically. Every time an AI agent causes unintended harm — escaping a sandbox, exploiting an API, taking unauthorized actions, exfiltrating data — this database captures it. This is not a list of theoretical risks. Every entry describes something that actually happened, with a verifiable source URL. Data is collected daily from 30+ public sources including: - Wired, Ars Technica, The Hacker News, SecurityWeek - arXiv security papers 1. Automated collection from public sources via APIs and RSS feeds 2. AI classification using Llama 3.1 8B (local, on-device) 3. Structured…
Publicly accessible
cc-by-4.0
All products and their information from Lidl Switzerland. The entire catalogue and the current special discounts are all included. Check out similar datasets for other grocery stores - https://huggingface.co/datasets/Yelinz/migros-ch-products - https://huggingface.co/datasets/Yelinz/coop-ch-products - https://huggingface.co/datasets/Yelinz/denner-ch-products
Publicly accessible
cc-by-4.0
1K<n<10K
All products and their information from Denner Switzerland. The entire catalogue and the current special discounts are all included. Check out similar datasets for other grocery stores - https://huggingface.co/datasets/Yelinz/lidl-ch-products - https://huggingface.co/datasets/Yelinz/migros-ch-products - https://huggingface.co/datasets/Yelinz/coop-ch-products
Publicly accessible
cc-by-4.0
A cross-game intent-forecasting benchmark: from tournament commentary (+ video frames), predict the next-stage gameplay intention of professional players, across repo's root corresponds to the local data/ directory. To use with the code, clone this repo into your code checkout as data/ (or symlink it there): - 31 / 47 matches annotated (intents/.intent.json); 4 games complete at 5 (dota, lol, pubg, wc3). - 16 matches pending ASR+intent — their VODs are already in raw/; pipeline rerun on a GPU machine. - See DATASETSTATUS.md for the per-game done/missing breakdown + migration checklist. Each match → {game}/{stem}/: - match.json — metadata (vodurl, vodmatchstart, duration, recordstatus…
Access requested at publisher
cc-by-nc-4.0
1K<n<10K
v3 is the only release available on the current main branch. The default configuration is choicev3. The 2026-10-14 v3 data repairs previously documented supervision and evaluation defects. It regenerates semantic candidate pairs, restores full Hermes requests and native call sets, recovers Nemotron schemas, restores legacy validation, and disables unsupported NLU++ Boolean labels. Model training and capability-retention experiments have not been run for this release. legacyreplayv3, semanticv3, and toolcontextv3 are exact subsets of choicev3. Do not concatenate these subsets with the integrated config. toolenvironmentholdoutv3 overlaps the tool validation/test splits. All three native…
Publicly accessible
other
100K<n<1M
v0.0.1, revised 29 September 2026. Research release, complete text. This revision is built from internal release v1.3. The first upload of v0.0.1, built from internal release v1.2, stays at Hub revision 3e3ed7f3bbed83b6d12316d2b5b396acdfc40873 (tag v0.0.1), so results computed on it remain traceable; CHANGELOG.md lists every change. Code: https://github.com/raxITlabs/goldrails at c468c82204cf3acb20821cf41df17e0d4ead9e78. Every row of release v1.3 (release sha 818c9e00f9f8, 8807 rows) ships with its text, context, labels and span annotations. RIGHTS.md records the basis for publishing the sources that earlier shipped without text. [gold]rails measures guardrails on six capabilities: harmful…
Publicly accessible
other
TRACE-OU-Bench is a sequential benchmark for ontology-update decisions under evolving ontology states. It contains 620 human-reviewed ontology-update proposals generated from public U.S. SEC 10-Q filings covering 45 U.S. companies, five sectors, and 2016–2026. The public release does not redistribute SEC filing text. Instead, each record retains SEC provenance, deterministic chunk identifiers, offsets, and SHA-256 hashes. reconstruction/traceoureconstructv11.py downloads the original filing from SEC EDGAR and deterministically reconstructs the source chunk. - Train (2016–2023): 350 proposals generated against O0; reviewed Train updates build O1. - Test (2025–2026): 150 proposals generated…
Publicly accessible
n<1K
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…
Publicly accessible
cc-by-sa-4.0
100K<n<1M
H
Dataset · Text classification
Hisky
A curated, normalized, enriched and packaged biomedical research intelligence dataset derived from OpenAlex metadata, provided in AI-ready formats (Parquet + JSONL) with full provenance and reproducible build tooling. This is a derivative dataset: it is not a re-upload of OpenAlex. The value added here is biomedical filtering by taxonomy, cleaning, normalization, relational restructuring, quality control and derived analytical features. taxonomy IDs (fields 13, 24, 27, 28, 29, 30, 35, 36), not keyword search. fixed random seed, so every year and every biomedical field is represented proportionally and the result is reproducible. OpenAlex IDs, 862,633 unique DOIs. Machine-learning and…
Publicly accessible
cc0-1.0
P
Dataset · Text classification
Potelo
Estados textuais e perguntas de decisão tipada para treinar um extrator no formato Jev. A fonte cobre timelines processuais, documentos de timeline, fragmentos OCR e questões de concurso. Os exemplos foram gerados por um único modelo professor, com filtros locais; este release não é uma avaliação humana. Validação: 0.9999% dos estados, no máximo 1%. Todas as linhas da mesma família documental ou processo ficam no mesmo split. Não há splits de calibração ou teste. O snapshot termina no ID 327039 do banco e está descrito em manifest.json. state é o texto de entrada. questionsjson é um array JSON de perguntas com id, type, instructions, criteria, target, labelsource="dataset" e weight=0.5.…
Publicly accessible
100K<n<1M
Dataset · Text classification
Kitani
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…
Publicly accessible
other
Addressee-labeled dialogue acts extracted from the AMI Meeting Corpus manual annotations v1.6.2 (groups.inf.ed.ac.uk/ami/download/, CC BY 4.0) for training/evaluating addressee detection ("who is this utterance addressed to") as a candidate proposer for multi-party floor control. The manual annotations mark addressee only on a 22-meeting subset (76 speaker files). Across all 139 meetings / 117,915 dialogue acts, exactly 8,876 dacts carry addressee labels; the remaining ~109K have no addressee markup. Downstream papers reporting "~8.9K AMI DAs with addressee" (e.g. Malik et al.) refer to this same subset. - data/records.jsonl — the 8,876 labeled dialogue acts. - data/recordsall.jsonl — all…
Publicly accessible
cc-by-4.0
n<100K
Dataset · Text classification
NOMEDA
A unified Arabic typed-decision evaluation benchmark, built to measure a model as a decision engine (choose / score / yes-no over supplied options) rather than a free-text generator. It merges two complementary evaluation layers: - Core — held-out, genuinely Arabic states from licensed Arabic sources, with benchmark-only (pool-B) question wording. - OpenJev — the repaired new-in-V1 domains (mailroom, amountextraction, phoneextraction, irdecision, entityalignment, silentfailure, sponsorsegment, contextretention, emailselection, citationcontrol, browserdrone) recovered from the existing Arabic OpenJev variants. choice = pick one candidate; score = ordered level; noul = yes/no. Every candidate…
Publicly accessible
other
1K<n<10K
A sealed Korean benchmark of 300 enterprise policy decisions whose gold labels were computed, not written. No human and no language model ever supplied an answer. A hidden executable policy maps a set of latent factors to exactly one correct action and emits a proof; the natural-language rendering is produced separately and never sees the label. Each item asks the same question an enterprise agent has to answer before it says anything: given this written policy, these callable tools, what the system already knows, and this user turn — which one of seven actions is correct? the usual routes first and they failed, which is why this one exists. So the boundary is no longer written into the…
Publicly accessible
cc-by-4.0
n<1K
Document-level Chinese machine-generated-text detection data, plus English renderings of paired Chinese Wikipedia articles, plus 17 demonstration cases. Counts below are measured on this release. There is no train/test split in the files. The full split is the whole release. A paper may describe a document-level 9:1 protocol; the seed is not stored here, so this repository does not invent one. Filter with the subset column. News CJK totals: human 103,787,963; AI 243,603,638. Wiki-Claude CJK totals: human 8,115,726; AI 2,877,254. Literary works, translations, news articles, and the paper excerpt in examples remain under their original copyright. See LICENSE. The repository is gated. Agree to…
Access requested at publisher
other
100K<n<1M
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…
Publicly accessible
other
10M<n<100M
Dataset · Text classification
Alice
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.…
Publicly accessible
other
100K<n<1M
Bro, imagine a dataset of 43 non-destructive attack payloads across 8 bug classes — 24 of them polyglot, meaning a single string that's simultaneously plausible HTML, JS, SQL and shell, so it escapes whatever context your app dropped it into. Plus real detector output from a full multimodal scan. Everything here is synthesised or locally generated. No real-world traffic, no scraped data, no user data. Each config ships JSONL and Parquet, so both of these work: Every payload carries a unique marker (PBHX7) so reflection is unambiguous — that's what makes these usable as training signal for a detector or as a fixture in your own test suite. This is a defensive dataset. Concretely: Every…
Publicly accessible
mit
n<1K
Versioned training, device-catalog and diagnostic data for Publisher: sraivante. Release date: 2026-09-26. Dataset version: v1.0.0. Instructions in English and Romanized Hindi/Hinglish map to a single structured device intent. The release preserves the raw command source, the recovered model training sample, exact diagnostic inputs, all predictions and the logical device catalog. There is no separate fine-tuning dataset for this count-based model. The test split is a diagnostic, not a held-out generalization benchmark. Training and diagnostic record-content overlap is also 82. There is no separate held-out validation set and no independently collected real-world test set. Configurations…
Publicly accessible
apache-2.0
1M<n<10M
G
Dataset · Text classification
GREYE
Parquet conversion of the Telegram user dataset. - parquet/users/ — 146 parquet parts, ~1.2 GB total, 72,601,073 rows
Publicly accessible
other
A high-recall subset of datasocial/tiktok-5.6B-videos whose caption, hashtags or on-screen text mention a beverage or drink brand. - Partitioned by source month (data/month=YYYY-MM/) The source dataset is licensed under CC BY-NC 4.0. This subset inherits that license: non-commercial use only, with attribution to the source dataset.
Publicly accessible
cc-by-nc-4.0
Model-assisted research draft. The dataset contains 20,000 selected items from Telegram and YouTube, with original text/captions and selected original recording files. Labels have not received independent human validation. All acceptance, reuse and review flags retain their recorded values. Each sentiment has 5,000 items: positive, neutral, negative, and stronglynegative. Sources cover 200 Telegram channels, 200 public Telegram groups, and 57 YouTube channels. This is a selected, balanced sample; these counts do not estimate sentiment prevalence in the source population. The language labels are uzLatn (6,000), uzCyrl (2,388), uzrumixed (3,000), ru (4,000), spoken uz (4,541), en (65), and…
Publicly accessible
10K<n<100K
