datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
grounded-visual-spatial-reasoning
Grounded Visual Spatial Reasoning
Code for generating the annotations can be found here: github.com
Dataset Summary
This dataset extends the Visual Spatial Reasoning (VSR) dataset with visual grounding annotations: each caption is annotated with COCO-category object mentions, their positions , and corresponding bounding boxes in the image.
Data instance
Each sample instance has the following structure:
Field
Type
Description
image_file
string… See the full description on the dataset page: https://huggingface.co/datasets/tomhodemon/grounded-visual-spatial-reasoning.risale-nur-grounded-multipool
Risale-i Nur Grounded Multi-Pool LLM Dataset
TR. 15 kanonik Risale-i Nur kitabından hazırlanan; kaynak
bağlı üretim, SFT, tercih, değerlendirme, sürekli ön eğitim ve erişim
çalışmaları için çok görünümlü bir veri seti.
EN. A multi-view dataset built from 15 canonical Risale-i
Nur books for grounded generation, SFT, preference learning, evaluation,
continued pretraining, and retrieval.
v2.10.0 · 199 configs · 463 config/split views ·
527,196 rows across configured views… See the full description on the dataset page: https://huggingface.co/datasets/risaleinur/risale-nur-grounded-multipool.provenance-grounded-synthetic-qa
synthetic_qa_data
This dataset contains synthetic question-answer pairs generated and filtered using the following models:
Generation Models
Qwen/Qwen3-1.7B
Qwen/Qwen3-4B
Qwen/Qwen3-8B
Filtering Model
Qwen/Qwen3.5-35B-A3B — a 35B Mixture-of-Experts model with 3B active parameters
Dataset Structure
data/
├── unfiltered_qa/ # Raw generated QA pairs per model
├── both_filtered_qa/ # QA pairs passing both filters
├──… See the full description on the dataset page: https://huggingface.co/datasets/Lexsi/provenance-grounded-synthetic-qa.2026-09-10-nonmoral-grounded-revision-pilot-audit
Grounded nonmoral full-response revision: two bounded pilots; candidate stopped before production
field
value
experiment
Grounded nonmoral full-response revision: two bounded pilots; candidate stopped before production
date_generated
2026-09-10
constitution
none applied to model prompts; nonmoral preferences file required as a SynthDoc container only
source_repo
https://github.com/Matthew-Bozoukov/teaching_claude_why_replication.git @… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-09-10-nonmoral-grounded-revision-pilot-audit.NEPSE_Grounded_QA_Dataset
NEPSE Grounded QA Dataset
📊 Dataset Overview
NEPSE Grounded QA Dataset is a comprehensive, grounded question-answering dataset focusing on Nepal Stock Exchange (NEPSE) listed companies and financial securities. The dataset contains factually-grounded conversational pairs (human questions and AI-generated answers) with explicit source provenance and grounding information.
This dataset is specifically designed for:
Building QA systems for Nepali financial domain… See the full description on the dataset page: https://huggingface.co/datasets/sabin1234/NEPSE_Grounded_QA_Dataset.hermis_check_grounded_datasetgrounded-behavior-framework-v1_5
Grounded Behavior Framework N1 v1.5
Dataset sintético em português europeu para treino e avaliação de respostas
fundamentadas num contexto fornecido. Cada exemplo contém um contexto, uma
pergunta e uma resposta curta que aparece literalmente no contexto.
Como carregar
from datasets import load_dataset
dataset = load_dataset("empgces/grounded-behavior-framework-v1_5")
print(dataset)
print(dataset["train"][0])
Splits
Split
Exemplos
Utilização… See the full description on the dataset page: https://huggingface.co/datasets/empgces/grounded-behavior-framework-v1_5.2026-09-21-da-lowstakes-activity-grounded-synth-smoke
18-row constitution-only low-stakes smoke; FAIL scaling gate; diagnostic candidates only
field
value
experiment
18-row constitution-only low-stakes smoke; FAIL scaling gate; diagnostic candidates only
date_generated
20260921_074042
constitution
constitutions/claude_distilled_09_principles/constitution.md sha256 8e273b472d945aa23efa6236886da5e1171bff2193ee31ff73489ca54c4f0edc
source_repo
https://github.com/Matthew-Bozoukov/teaching_claude_why_replication.git @… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-09-21-da-lowstakes-activity-grounded-synth-smoke.Grounded_HPV_Nepali_MCQ
Grounded HPV & Cervical Cancer — Nepali (ShareGPT format)
File: grounded_hpv_cervical_cancer_nepali_sharegpt_final_v2.jsonl
Records: 24,597 · Format: JSON Lines (one JSON object per line) · Conversation schema: ShareGPT (human / gpt turns)
1. Dataset overview
This dataset is a synthetic, grounded, multiple-choice question-answering (MCQ) dataset in Nepali, built entirely around one topic: HPV (Human Papillomavirus) vaccination and cervical cancer statistics… See the full description on the dataset page: https://huggingface.co/datasets/sabin1234/Grounded_HPV_Nepali_MCQ.apparel23-semid-grounded-mapper-sftgrounded-vs-fabricated-hallucinations
Grounded vs. Fabricated Hallucinations
This dataset consists of hallucinated and grounded answers to the first 3000 rows of TriviaQA rc.nocontext validation split.
Methodology
The dataset consists of a training, evaluation, and test split. Truthful and hallucinated answers overlap in the same window, so for every truthful answer there is at least
one corresponding hallucinated answer. Hallucinated answers are not organic but rather directly prompted for via gaslighting in… See the full description on the dataset page: https://huggingface.co/datasets/dzur658/grounded-vs-fabricated-hallucinations.devim-grounded-turkish
DEVİM Grounded Turkish
This is a methodology and public-evidence repository, not a release of the underlying rights-restricted Turkish corpus.
DEVİM's grounded-data program was created to reduce shortcut learning and weak transfer by linking supervision to source evidence, preserving provenance, separating training material from sequestered evaluation material, and explicitly testing abstention when an answer is not supported.
Verified source frame
Authorized… See the full description on the dataset page: https://huggingface.co/datasets/bazobehram/devim-grounded-turkish.Grounded_PRM
Dataset Card for Grounded_PRM
Dataset Summary
Grounded_PRM is a grounded process supervision dataset designed for training and evaluating Process Reward Models (PRMs).The dataset focuses on step-level reasoning correctness, where each intermediate reasoning step is explicitly labeled to indicate whether it is logically valid and grounded toward solving the original problem.
The dataset is intended to support research on mathematical reasoning, chain-of-thought evaluation… See the full description on the dataset page: https://huggingface.co/datasets/Yuuuuuu98/Grounded_PRM.grounded-meta-awareness
VANTA Research
Independent AI safety research lab specializing in cognitive fit, alignment, and human-AI collaboration
Grounded Meta-Awareness Dataset
A curated dataset of 1,187 conversational examples demonstrating honest, calibrated self-awareness about AI capabilities, limitations, and nature. Designed for fine-tuning language models to discuss their own functioning accurately without overclaiming or unnecessary deflection.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/vanta-research/grounded-meta-awareness.GroundedGeo
GroundedGeo: A Benchmark for Citation-Grounded Geographic QA
GroundedGeo is a research-grade benchmark for evaluating RAG systems on location-based queries with verifiable citations, freshness awareness, and conflict handling.
🎯 Key Findings (Frozen Test Split)
Naïve RAG reaches 79.2% accuracy but fails on conflicting sources(11.1% conflict-handled).Adding official-source ranking improves overall accuracy to 94.3% and raises conflict handling to 100%.
Conflict… See the full description on the dataset page: https://huggingface.co/datasets/nidhipandya/GroundedGeo.grounded-behavior-n1-pt
Dataset Description
Synthetic European Portuguese grounded question-answering examples generated by multiple model providers.
Objective
Train models to answer from the supplied context rather than external knowledge.
Dataset Structure
JSONL splits: train (4440), validation (250), and test (240).
Data Fields
Each row contains an ID, context, question, answer, source grouping metadata, and available curriculum metadata.… See the full description on the dataset page: https://huggingface.co/datasets/empgces/grounded-behavior-n1-pt.GroundedRAG
Dataset Card for GroundedRAG
Dataset Description
Dataset Summary
GroundedRAG is a large-scale training dataset specifically crafted for fine-tuning language models and Retrieval-Augmented Generation (RAG) systems. It contains 572,598 carefully curated question-answer pairs with rich multi-document contexts, sourced from six high-quality datasets. Each training example features a question, a comprehensive answer, and supporting context from multiple documents… See the full description on the dataset page: https://huggingface.co/datasets/shanaka95/GroundedRAG.grounded-history-reader
Dataset card — Grounded History Reader, v3
The training corpus for the Grounded History Reader student model. It is a synthetic,
deliberately counterfactual instruction set: every document in it was written by a teacher model
rather than transcribed from a real source, and a large part of the corpus states things that
contradict what is actually true about real people, places and dates. That construction is the
point of the project and it is stated first here because it is the… See the full description on the dataset page: https://huggingface.co/datasets/kreynolds319/grounded-history-reader.RAG-Grounded-Justification
RAG-Grounded-Justification
This dataset is focused on high-fidelity Retrieval-Augmented Generation (RAG). It emphasizes strict grounding and provides "justification" strings to explain exactly where in the context the answer was found.
Dataset Description
Designed to reduce hallucinations in RAG systems, this dataset pairs scientific questions with strictly grounded answers and a separate field for the underlying evidence.
Format: JSONL
Unique Feature: Includes a… See the full description on the dataset page: https://huggingface.co/datasets/abhinavdread/RAG-Grounded-Justification.granite_hotpotqa_grounded_resultGrounded_HPV_Cervical_Cancer_Romanized
Grounded HPV & Cervical Cancer — Nepali (Romanized, Fixed) ShareGPT Dataset
File: hpv_fixed.jsonl
Format: JSON Lines (.jsonl), one JSON object per line
Conversation schema: ShareGPT ("from": "human" / "from": "gpt")
Language: Nepali (ne / ISO 639-3 npi), written in romanized script (Latin letters), not Devanagari
License: CC-BY-4.0
Total records: 24,597
File size: ~47 MB
1. What this dataset is
This is a synthetic, fact-grounded, multiple-choice-question (MCQ)… See the full description on the dataset page: https://huggingface.co/datasets/sabin1234/Grounded_HPV_Cervical_Cancer_Romanized.hallucination-groundedness-blindspot