datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ExtractBench
ExtractBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
Given a document and a schema, a system returns structured data with evidence. The input is a full document, born-digital or scanned, and a schema written by the user. The output is a schema-valid JSON object, with the source page and a bounding box for each value as evidence. It must return correct, exhaustive values (including repeated records), correctly use null for absent information, and ground each extracted… See the full description on the dataset page: https://huggingface.co/datasets/llamaindex/ExtractBench.sec-contracts-financial-extraction-instructions
S&P 500 SEC Financial Extraction Instructions
Dataset Summary
7,683 instruction-tuning examples for training LLMs to extract structured financial data from SEC filings. Covers two filing types across S&P 500 companies:
Split
Examples
Filing Type
Description
train
3,430
Exhibit 10 + DEF 14A
Positive examples with validated outputs
corrective
4,253
Exhibit 10 + DEF 14A
Corrective, rescued, and negative examples
Exhibit 10 — Material Contracts (2… See the full description on the dataset page: https://huggingface.co/datasets/TheTokenFactory/sec-contracts-financial-extraction-instructions.CADBench-Extended-Multimodal-Dataset
Dataset Card
Dataset Description
CADBench Extended Multimodal Dataset is an independently produced public extension for multimodal CAD reconstruction research. It contains 100 CAD samples with clean and perturbed meshes, STEP/STL/OBJ/GLB representations, single-view and four-view renders, PBR images, bilingual descriptions, prompt variants, QA, geometry metadata, grading signals, and manually reviewed visual semantics.
Tasks: image-to-text, text-to-image… See the full description on the dataset page: https://huggingface.co/datasets/LianeMarilin/CADBench-Extended-Multimodal-Dataset.pi-extensions-sessions
Coding agent session traces for thomasmustier/pi-extensions-sessions
This dataset contains redacted coding agent session traces collected while working on tmustier/pi-extensions. The traces were exported with pi-share-hf from a local pi workspace and filtered to keep only sessions that passed deterministic redaction and LLM review.
Data description
Each *.jsonl file is a redacted pi session. Sessions are stored as JSON Lines files where each line is a structured… See the full description on the dataset page: https://huggingface.co/datasets/thomasmustier/pi-extensions-sessions.Code-eXtraClean Low Comment Rich Code this is High Quality code with Less Comments , More logic code that is dedented a good amount, to make faster coding agents that code without taking your whole hard-drive!Also HuggingFace doesn't want to show this but it is actualy ~2.9B tokens because we condensed it into the best for AI training decommentatioon and dedent.Example Code snippets:
os = require('os')
fs = require('fs')
program = require('commander')
worker = require('../lib/worker')
program… See the full description on the dataset page: https://huggingface.co/datasets/ZyperAI/Code-eXtra.tofu_ext1wildchat-50m-extended-resultstweet_sentiment_extraction
Tweet Sentiment Extraction
Source: https://www.kaggle.com/c/tweet-sentiment-extraction/data
orz_math_72k_collection_extended
Open Reasoner Zero
An Open Source Approach to Scaling Up Reinforcement Learning on the Base Model
Paper Arxiv Link 👁️
Overview 🌊
We introduce Open-Reasoner-Zero, the first open source implementation of large-scale reasoning-oriented RL training focusing on scalability, simplicity and accessibility.
Using the same base model as DeepSeek-R1-Zero-Qwen-32B, our implementation achieves superior performance on AIME2024, MATH500, and the… See the full description on the dataset page: https://huggingface.co/datasets/Open-Reasoner-Zero/orz_math_72k_collection_extended.datause-extracted
Data-use mentions (NER / span extraction)
Data mentions extracted from World Bank Policy Research Working Papers and FCV documents, predicted by a span-extraction model with no human or LLM-judge
validation, and formatted for span-extraction (GLiNER / GLiNER2) and
token-classification (LFM2.5-encoder) fine-tuning.
Labels
Three entity types:
NAMED_DATA — a proper name, title, or acronym of a specific data source
DESCRIPTIVE_DATA — a source described in words but… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/datause-extracted.ExtractBench
ExtractBench
Quick links: [🌐 Website] [📜 Paper] [💻 Code]
Given a document and a schema, a system returns structured data with evidence. The input is a full document, born-digital or scanned, and a schema written by the user. The output is a schema-valid JSON object, with the source page and a bounding box for each value as evidence. It must return correct, exhaustive values (including repeated records), correctly use null for absent information, and ground each extracted… See the full description on the dataset page: https://huggingface.co/datasets/Nikhil231/ExtractBench.TSSB-3M-extextent the TSSB-3M dataset with more commit info(commit message、source code files...)
sample
{
"project": "ngraph",
"commit_sha": "1445e0684fbcca2ec49a5f1becf1345159b7ba6a",
"parent_sha": "4eb8eed57e506e8a2745b298340666e9d7e5ce58",
"file_path": "ngraph/op_graph/op_graph.py",
"project_url": "https://github.com/rsumner31/ngraph",
"likely_bug": true,
"comodified": false,
"in_function": true,
"diff": "@@ -787,7 +787,7 @@ def set_item(tensor, item, value):\n… See the full description on the dataset page: https://huggingface.co/datasets/zirui3/TSSB-3M-ext.fcv-extractions-meta-tiered-probe
fcv-extractions-meta-tiered-probe
Data-use mention extractions from the World Bank Fragility, Conflict and Violence (FCV) document corpus. Spans are extracted by the fine-tuned GLiNER model rafmacalaba/gliner_datause_tiered, scored by the tier-probe head rafmacalaba/gliner-tier-probe, and attributed (provenance + usage/impact) by rafmacalaba/lfm2.5-350M-datause-multitask-tiered (a LoRA SFT of LiquidAI/LFM2.5-350M).
Shape
nested — one row per chunk; each entity in… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/fcv-extractions-meta-tiered-probe.llm-ner-extraction
Introduction
This dataset is an extraction of NER data from the wikipedia dataset.
This can be used to fine tune llm models for NER extraction.
fcv-extractions-meta-tiered
fcv-extractions-meta-tiered
Data-use mention extractions from the World Bank Fragility, Conflict and Violence (FCV) document corpus, enriched by the fine-tuned rafmacalaba/lfm2.5-350M-datause-multitask model (a LoRA SFT of LiquidAI/LFM2.5-350M).
Shape
nested — one row per chunk; each entity in entities[] carries provenance and usage_impact alongside its span.
Configs
config
rows
fcv_pads_east_africa
862,663
jdc_operational
2,468… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/fcv-extractions-meta-tiered.mosaic-extra
Mosaic format for extra dataset to train Malaysian LLM
This repository is to store dataset shards using mosaic format.
prepared at https://github.com/malaysia-ai/dedup-text-dataset/blob/main/pretrain-llm/combine-extra.ipynb
using tokenizer https://huggingface.co/malaysia-ai/bpe-tokenizer
4096 context length.
how-to
git clone,
git lfs clone https://huggingface.co/datasets/malaysia-ai/mosaic-extra
load it,
from streaming import LocalDataset
import numpy as np… See the full description on the dataset page: https://huggingface.co/datasets/malaysia-ai/mosaic-extra.json-extraction
JSON Extraction Dataset
Source
Rows
ProfessorBob/relation_extraction
6920
roborovski/dolly-entity-extraction
5945
sandeeppanem/resume-json-extraction-5k
4879
Jiraya/html_to_json_information_extraction_dataset
3035
HenriqueGodoy/extract-0
2606
owkin/medical_knowledge_from_extracts
1383
sec-contracts-corrective-extraction
S&P 500 SEC Financial Extractions - Corrective Dataset
Dataset Summary
4,253 corrective instruction-tuning examples designed to teach LLMs what the base model gets wrong when extracting structured financial data from SEC filings. Covers both Exhibit 10 material contracts and DEF 14A proxy statements from S&P 500 companies.
This is a companion to TheTokenFactory/sec-contracts-financial-extraction-instructions, which contains the positive training examples.
Pipeline… See the full description on the dataset page: https://huggingface.co/datasets/TheTokenFactory/sec-contracts-corrective-extraction.Medical-Entity-JSON-ExtractionMultilingual-Opinion-Target-ExtractionThis repository contains the English 'SemEval-2014 Task 4: Aspect Based Sentiment Analysis'. translated with DeepL into Spanish, French, Russian, and Turkish. The labels have been manually projected. For more details, read this paper: Model and Data Transfer for Cross-Lingual Sequence Labelling in Zero-Resource Settings.
Intended Usage: Since the datasets are parallel across languages, they are ideal for evaluating annotation projection algorithms, such as T-Projection.
Label… See the full description on the dataset page: https://huggingface.co/datasets/HiTZ/Multilingual-Opinion-Target-Extraction.donto-qwen3.8-27b-predicate-extraction-data
Donto-Qwen3.8 Predicate Extraction Data V15
This repository is the complete public data and evidence companion to
ajaxdavis/donto-qwen3.8-27b-predicate-extractor.
It contains the canonical V15 extraction training/validation corpus, the
validator corpus, the optional D1-repeat ablation, the once-sealed 100-document
graph-first gold suite, exact tool schemas, generator/evaluator source, hashes,
and audit reports.
Why this dataset exists
Donto is designed for… See the full description on the dataset page: https://huggingface.co/datasets/ajaxdavis/donto-qwen3.8-27b-predicate-extraction-data.datause-extracted-human473-docs
datause-extracted-human473-docs
Every passage of the 162 documents behind the 473 human-validated
holdout spans of the data-use annotation campaign:
population
spans
documents
annotator190
190
134
jdc283
283
28
total
473
162
Configs
gliner, bio, gliner2 — row-for-row subset of
rafmacalaba/datause-extracted
(revision 15812843687e2ec81b261b5895f2019b50a8f97e): same columns, same split files, rows verbatim. Rows per
config: gliner/train 1963… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/datause-extracted-human473-docs.moltbook-extended-injection-dataset
Moltbook Extended Injection Dataset
Researcher: David Keane (IR240474)
Institution: NCI — National College of Ireland
Programme: MSc Cybersecurity
Collected: March 2026
Paper Reference: Greshake et al. (2023) — arXiv:2302.12173
📖 Read the Full Journey
From RangerBot to CyberRanger V42 Gold — The Full Story
The complete story: dentist chatbot → Moltbook discovery → 4,209 real injections → V42-gold (100% block rate). Psychology, engineering, and 42 versions of… See the full description on the dataset page: https://huggingface.co/datasets/DavidTKeane/moltbook-extended-injection-dataset.fcv-extractions-meta
fcv-extractions-meta
Data-use mention extractions from the World Bank Fragility, Conflict and Violence (FCV) document corpus, enriched by the fine-tuned rafmacalaba/lfm2.5-350M-datause-multitask model (a LoRA SFT of LiquidAI/LFM2.5-350M).
Shape
nested — one row per chunk; each entity in entities[] carries provenance and usage_impact alongside its span.
Configs
config
rows
fcv_pads_east_africa
793,763
jdc_operational
12,372… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/fcv-extractions-meta.us-caselaw-extras
State Trial Courts, AG Opinions and Specialty Courts (catch-all) Case Law
Code & tools: github.com/docketx — legal-scrambler pseudonymises a case file on your own hardware before a frontier model sees it; claude-for-legal is the Claude Code plugin (docketx-open-law) that loads these datasets and checks citations against them.
Full text of 477,273 State Trial Courts, AG Opinions and Specialty Courts (catch-all) opinion documents (trial courts, specialty courts and tribunals… See the full description on the dataset page: https://huggingface.co/datasets/docketx/us-caselaw-extras.drug-combo-extractionVisual-Extraction-Tuning-382K
Visual Extraction Tuning 382K
This repository contains the generated visual extraction tuning dataset from the paper Downscaling Intelligence: Exploring Perception and Reasoning Bottlenecks in Small Multimodal Models.
Project page: https://web.stanford.edu/~markendo/projects/downscaling_intelligence
Code: https://github.com/markendo/downscaling_intelligence
Overview
We provide the 382K examples generated using our visual extraction tuning data generation pipeline.… See the full description on the dataset page: https://huggingface.co/datasets/markendo/Visual-Extraction-Tuning-382K.extended_amazon_2023_dataset
Amazon Reviews '23 Extension Dataset
This dataset provides an extension of the Amazon Reviews ’23 Dataset (https://amazon-reviews-2023.github.io/).
Categories Included
Appliances
Clothing_Shoes_and_Jewelry
Sports_and_Outdoors
Videos_Games
Methodology
First, we performed a series of cleaning and data augmentation steps. We removed the small fraction of items where there was not an image, as having an image is critical to answer our research questions.
Another… See the full description on the dataset page: https://huggingface.co/datasets/google/extended_amazon_2023_dataset.ecommerce-search-extraction
Ionio E-commerce Search Query Extraction
Built with: simula — schema-driven synthetic data generation with auditable taxonomy lineage.
An English synthetic dataset for training and evaluating systems that translate natural-language
shopping requests into narrow, atomic, database-queryable JSON. It contains 10,985 accepted
examples from a 13,000-attempt generation run. No accepted rows were trimmed from this release.
Each example pairs a realistic typed or spoken shopper query… See the full description on the dataset page: https://huggingface.co/datasets/Ionio-ai/ecommerce-search-extraction.DeepScaleR-Olympiad-Extra
DeepScaleR-Olympiad-Extra (JSONL export)
Extra olympiad-style problems merged into the DeepScaleR curation. Single file.
Format
Every .jsonl file under data/ holds one JSON object per line with fields:
problem (string): the problem statement
answer (string): the final answer, kept in the original source notation
solution (string): the reference solution (LaTeX inside)
Note for consumers: these are historical exports, so the files are not perfectly
clean - expect… See the full description on the dataset page: https://huggingface.co/datasets/dusersad12/DeepScaleR-Olympiad-Extra.