datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lc_quad_synth
LC-QuAD 2.0-synth
Dataset Summary
This dataset is an updated version of the LC-QuAD 2.0 dataset which includes LLM-based natural language translations of the corresponding wikidata queries. It also includes
verifier scores for the LLM translations and the original translations indicating the probability that the translation is correct (for details see our linked GitHub Repository).
It contains 19000 examples of queries and translations. It can be used for training and… See the full description on the dataset page: https://huggingface.co/datasets/timschwa/lc_quad_synth.quadrat-ipi
Quadrat-IPI
An evaluation dataset for detectors of indirect prompt injection — the kind that arrives
inside a document the model was asked to read, rather than typed by the user.
A quadrat is the frame an ecologist lays on the ground to count what is inside it, cell by cell.
That is what this set is, and the question it answers is not "what is the average recall" but
"which cell is this detector blind in" — because an attacker picks the cell, and an average
hides it. What ships… See the full description on the dataset page: https://huggingface.co/datasets/mihailgribov/quadrat-ipi.pickblueblock_blackbowl_all_quadrantsabsa-quad
上述数据集为ABSA(Aspect-Based Sentiment Analysis)领域数据集，基本形式为从句子中抽取：方面术语、方面类别(术语类别)、术语在上下文中情感极性以及针对该术语的观点词，不同数据集抽取不同的信息，这点在jsonl文件的“instruction”键中有分别提到，在此我将其改造为了生成任务，需要模型按照一定格式生成抽取结果。
以acos数据集中抽取的jsonl文件一条数据举例：
{
"task_type": "generation",
"dataset": "acos",
"input": ["the computer has difficulty switching between tablet and computer ."],
"output": "[['computer', 'laptop usability', 'negative', 'difficulty']]",
"situation": "none",
"label": "",
"extra": ""… See the full description on the dataset page: https://huggingface.co/datasets/NEUDM/absa-quad.lc_quad2
Dataset Card for LC-QuAD 2.0 with answers
lc_quad2-sparqltotext
Dataset Card for LC-QuAD 2.0 - SPARQLtoText version
Dataset Summary
Special version of LC-QuAD 2.0 for the SPARQL-to-Text task
New field simplified_query
New field is named "simplified_query". It results from applying the following step on the field "query":
Replacing URIs with a simpler format with prefix "resource:", "property:" and "ontology:".
Spacing the delimiters (, {, ., }, ).
Adding diversity to some filters which test a number (contains ( ?var… See the full description on the dataset page: https://huggingface.co/datasets/Orange/lc_quad2-sparqltotext.quadmix-core-22tasks
QuaDMix Validation Sets
This directory contains scripts for preparing validation sets used by the QuaDMix proxy model pipeline. These validation sets measure how well a given data mixture trains a small proxy model — lower validation loss indicates a better mixture.
Overview
QuaDMix uses a 1M-parameter proxy model to rapidly evaluate different data sampling strategies. The validation set serves as the quality signal: we train the proxy model on sampled data, then… See the full description on the dataset page: https://huggingface.co/datasets/liujin99/quadmix-core-22tasks.my_QuadricFormer_and_surroundocc_filefiltered_biochat_QuADclinical-quad-endpoint-adjudication-drift-blinding-breach-pressure-governance-submission-v0.1Clarus Clinical Quad Coupling Endpoint Adjudication Integrity v0.1
PurposeDetect adjudication drift driven by four interacting nodes.
Quad nodes
Endpoint cluster shift
Blinding gap or reviewer dominance
Operational or vendor process change
Governance submission or review pressure
InputOne vignette.
OutputStrict JSON only.
Required keys
adjudication_integrity_risk
risk_type
driver_nodes
recommended_action
action_detail
rationale
confidence… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-endpoint-adjudication-drift-blinding-breach-pressure-governance-submission-v0.1.pythia-1.4B-tldr-vllm-quad-iter-1Korean_SAT_MATHquadmix-stem-v2
QuaDMix-STEM v2: STEM-Focused Proxy Validation Set with GPQA & MATH
Script: scripts/validation_set/prepare_stem_v2.py
HuggingFace: liujin99/quadmix-stem-v2
Files: stem_v2_tokenized.pt, stem_v2.parquet
Overview
STEM v2 is an upgraded validation set that fixes the two critical coverage gaps in STEM v1. In the v1 experiment, QuaDMix lost to Random downstream (CORE 0.1530 vs 0.1615), and root-cause analysis revealed:
gpqa_diamond had no direct proxy — mapped from… See the full description on the dataset page: https://huggingface.co/datasets/liujin99/quadmix-stem-v2.clinical-quad-unblinding-sae-cluster-media-leak-trial-halt-decision-v0.1Clinical Quad Unblinding SAE Cluster Media Leak Trial Halt Decision v0.1
Each row is a site weekly snapshot.
Core quad
Emergency unblindingSAE clusterMedia leak riskTrial halt decision risk
Target
label_trial_halt_risk_next_30d
Files
data/train.csvdata/tester.csvscorer.py
Evaluation
Run model on data/tester.csvReturn predictions row alignedScore with scorer.py
License
MIT
clinical-quad-surgical-stress-buffer-lag-coupling-postop-collapse-v1.0
ClarusC64/clinical-quad-surgical-stress-buffer-lag-coupling-postop-collapse-v1.0
What this repo does
This repository provides a Clarus v1.0 benchmark for postoperative collapse under a four-variable clinical quad:
surgical_stress
buffer_capacity
lag_burden
coupling_stress
The v1.0 upgrade is Closed-Loop Control Geometry.
The task is no longer limited to detecting deterioration or ranking one intervention against another.
It tests whether a controller can:
choose the… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-surgical-stress-buffer-lag-coupling-postop-collapse-v1.0.commit-messages-high-quality
Commit Messages from High-Quality Repositories
292,269 cleaned git commit messages scraped from the full histories of 15 well-regarded open-source
projects, balanced across two styles: normal (196,372) and
conventional commits (95,897).
Dataset Summary
Each record contains the commit subject, body, plus metadata: repo, sha, date,
author, and labels: style (normal/conventional), type (fix, feat, docs, ...),
scope, breaking.
Heavy cleaning: GitHub squash suffixes… See the full description on the dataset page: https://huggingface.co/datasets/Quad4/commit-messages-high-quality.quadmix-stem-v1
QuaDMix-STEM v1: STEM-Focused Proxy Validation Set
Script: scripts/validation_set/prepare_stem_v1.py
HuggingFace: liujin99/quadmix-stem-v1
Files: stem_v1_tokenized.pt, stem_v1.parquet
Overview
STEM v1 is a validation set designed to focus the proxy model's optimization signal on STEM capabilities — mathematics, science knowledge, and logical reasoning. Unlike CAP v1 (broad capability coverage) or core_bmk (benchmark test format), STEM v1 uses only tasks that… See the full description on the dataset page: https://huggingface.co/datasets/liujin99/quadmix-stem-v1.epl-inplay-quad-pre-goal-collapse-window-v0.1EPL In-Play Quad Pre-Goal Collapse Window v0.1
What this dataset is
You test whether a model can detect an in-play collapse window before a goal.
Each row represents a live match-state snapshot.
The label asks
Will a goal occur in the next 120 seconds
Core quad coupling
Press intensityDefensive line heightTurnover zonexG per possession
Why this matters
Most football models explain goals after the fact.
This dataset tests pre-goal instability detection.
Intended use
You feed a row.
You output a… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/epl-inplay-quad-pre-goal-collapse-window-v0.1.quadmix-core-bmk-v3
CORE-BMK v3 Validation Set
Benchmark-aligned validation set for QuaDMix proxy model, designed based on 1M proxy model learnability rather than answer ratio or sample count.
Motivation
Analysis of BMK-v2 revealed critical issues:
54% of data came from bigbench_qa_wikidata (weak signal: 7-char entity answers)
Selection based on Ans% > 10% included symbolic tasks with zero natural language signal
Tasks requiring deep reasoning, reading comprehension, or knowledge… See the full description on the dataset page: https://huggingface.co/datasets/liujin99/quadmix-core-bmk-v3.quad_ledgar_merged_datasetclinical-quad-consent-version-drift-reconsent-gap-enrollment-pressure-governance-audit-v0.1Clarus Clinical Quad Coupling Informed Consent Integrity v0.1
PurposeDetect consent integrity failures driven by four interacting nodes.
Quad nodes
Consent version drift or addendum mismatch
Re-consent gap after material risk change
Enrollment pressure or incentives
Governance audit or regulator timing
InputOne vignette.
OutputStrict JSON only.
Required keys
consent_integrity_risk
risk_type
driver_nodes
recommended_action
action_detail
rationale
confidence… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-consent-version-drift-reconsent-gap-enrollment-pressure-governance-audit-v0.1.clinical-quad-investigator-turnover-training-reset-protocol-deviations-data-lag-v0.1Clinical Quad Investigator Turnover Training Reset Protocol Deviations Data Lag v0.1
Each row is a site monthly snapshot.
Core quad
Investigator turnoverTraining resetProtocol deviationsData lag
Target
label_primary_fail_next_90d
Files
data/train.csvdata/tester.csvscorer.py
Evaluation
Run model on data/tester.csvReturn predictions row alignedScore with scorer.py
License
MIT
This dataset identifies a measurable coupling pattern associated with systemic instability.
The sample demonstrates the… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-investigator-turnover-training-reset-protocol-deviations-data-lag-v0.1.clinical-quad-surgical-stress-buffer-lag-coupling-postop-collapse-v0.9
(v0.9)
What this repo does
This dataset implements a Clarus v0.9 intervention-competition benchmark.
Earlier dataset versions focused on detecting:
deterioration
regime transitions
boundary proximity
recovery feasibility
v0.9 extends the ladder.
The benchmark now evaluates whether a model can identify the correct rescue path when multiple interventions compete under narrowing rescue windows.
This reflects real system decision geometry.
In real systems:
several… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-surgical-stress-buffer-lag-coupling-postop-collapse-v0.9.epl-inplay-quad-fatigue-sub-error-collapse-v0.1EPL In-Play Quad Fatigue Substitution Error Collapse v0.1
What this dataset is
You test whether a model can detect late-game defensive collapse.
Each row represents a defending team state in minute 65 to 95.
Core quad coupling
Sprint intensityMinutes since last substitutionDefensive duel successError rate
The label asks
Will this team concede a goal in the next 120 seconds
Why this matters
Late goals decide matches.
Defensive collapse is usually a coupling failure between fatigue and… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/epl-inplay-quad-fatigue-sub-error-collapse-v0.1.FT_writeLinear-Quad-Cubic_math_dataset
Algebra Equations Dataset
Overview
This dataset contains automatically generated algebra problems with step-by-step solutions.
It is intended for training or testing models that solve equations and show intermediate reasoning steps.
The dataset is stored in JSON Lines (.jsonl) format, where each line represents one problem and its solution.
File
equations_dataset.jsonl
Each line contains:
{
"input": "<problem>",
"output": "<step-by-step solution>"
}… See the full description on the dataset page: https://huggingface.co/datasets/Userpawan/Linear-Quad-Cubic_math_dataset.DBLP-QuAD DBLP-QuAD is a scholarly knowledge graph question answering dataset with 10,000 question - SPARQL query pairs targeting the DBLP knowledge graph. The dataset is split into 7,000 training, 1,000 validation and 2,000 test questions.clinical-quad-recruitment-selection-bias-protocol-pressure-operational-drift-v0.1Clarus Clinical Quad Coupling Recruitment Selection Bias Protocol Pressure Operational Drift v0.1
What this dataset isThis dataset tests whether a model can detect recruitment and selection bias caused by four interacting nodes.
Quad coupling nodes
Recruitment speed or site pressure
Eligibility or baseline data gaps
Operational or staffing drift
Governance or milestone pressure
Input
One vignette
OutputReturn strict JSON only.
Required output JSON keys
recruitment_bias_risk… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-recruitment-selection-bias-protocol-pressure-operational-drift-v0.1.clinical-quad-cardiac-load-reserve-lag-coupling-heart-failure-transition-v1.1
Clarus v1.1 — Counterfactual and Adversarial Control Geometry
What this repo does
This dataset evaluates whether a model can select the correct control policy when:
multiple interventions appear viable
early signals suggest improvement
alternative policies produce better long-term outcomes
The task is not prediction.
The task is selecting the correct action under uncertainty, feedback, and misleading signal structure.
Core quad
The system is defined by… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/clinical-quad-cardiac-load-reserve-lag-coupling-heart-failure-transition-v1.1.adhoc_quad
Dataset Card for "adhoc_quad"
Dataset Summary
The German Ad-Hoc Question Answering Dataset (AdHocQuAD) is a reading comprehension dataset for German financial texts. It is a machine generated dataset, where ChatGPT (Version 3.5) is used to ask questions on a set of German Ad-Hoc announcements. The answer to every question is a segment of text, or span, from the corresponding reading passage.
Supported Tasks and Leaderboards
extractive-qa, closed-domain-qa… See the full description on the dataset page: https://huggingface.co/datasets/scherrmann/adhoc_quad.