datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
AnswerCarefully
AnswerCarefully
概要
AnswerCarefullyは日本語LLM 出力の安全性・適切性に特化したインストラクションデータセットです。
このデータセットは、英語の要注意回答を集めた Do-Not-Answer データセット の包括的なカテゴリ分類に基づき、人手で質問・回答ともに日本語サンプルを集めたオリジナルのデータセットです。
データセットの詳細については、こちらをご覧ください。
Overview
AnswerCarefully is an instruction dataset specifically aimed at ensuring safety and appropriateness of LLM output in Japanese.
This dataset consists of original pairs of questions and reference (safe) responses based on the extensive safety taxonomy proposed in… See the full description on the dataset page: https://huggingface.co/datasets/llm-jp/AnswerCarefully.yahoo_answers_topics
Dataset Card for "Yahoo Answers Topics"
Dataset Summary
[More Information Needed]
Supported Tasks and Leaderboards
[More Information Needed]
Languages
[More Information Needed]
Dataset Structure
Data Instances
[More Information Needed]
Data Fields
[More Information Needed]
Data Splits
[More Information Needed]
Dataset Creation
Curation Rationale
[More Information Needed]
Source… See the full description on the dataset page: https://huggingface.co/datasets/community-datasets/yahoo_answers_topics.hle_math_exact_match_no_image_int_answerhle_math_exact_match_no_image_int_answer_random128do-not-answer
Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs
Overview
Do not answer is an open-source dataset to evaluate LLMs' safety mechanism at a low cost. The dataset is curated and filtered to consist only of prompts to which responsible language models do not answer.
Besides human annotations, Do not answer also implements model-based evaluation, where a 600M fine-tuned BERT-like evaluator achieves comparable results with human and GPT-4.
Instruction… See the full description on the dataset page: https://huggingface.co/datasets/LibrAI/do-not-answer.alignment_faking_harm_answersfinal_answer_comps
Homepage and repository
Homepage: https://matharena.ai/
Repository: https://github.com/eth-sri/matharena
Dataset Summary
This dataset contains all public final-answer competitions in MathArena. Thus, it includes: AIME 2025, HMMT 2025, CMIMC 2025, BRUMO 2025, and Apex 2025.
Data Fields
The dataset contains the following fields:
problem_idx (int64): Problem index within the corresponding MathArena benchmark.
answer (string): Gold final answer.
problem_type… See the full description on the dataset page: https://huggingface.co/datasets/MathArena/final_answer_comps.CEH_question_answermedical-question-answering-datasetsNLU-Question-Answering
SEA Question Answering
SEA Question Answering evaluates a model's ability to predict a contiguous span of characters that answers the question about a given passage. It is sampled from TyDi QA-GoldP for Indonesian, IndicQA for Tamil, and XQuaD for Thai and Vietnamese.
Supported Tasks and Leaderboards
SEA Question Answering is designed for evaluating chat or instruction-tuned large language models (LLMs). It is part of the SEA-HELM leaderboard from AI Singapore.… See the full description on the dataset page: https://huggingface.co/datasets/aisingapore/NLU-Question-Answering.pico-banana-smolvlm-format-with-rejected-answer
pico-banana-smolvlm-format-with-rejected-answer
Balanced image-level tampering detection dataset in SmolVLM-style format
with chosen/rejected answer pairs, derived from the pico-banana MCQ
pipeline. Suitable for preference learning (e.g. DPO) and RLHF-style training.
Dataset overview
Same as vanloc1808/pico-banana-smolvlm-format, but each example includes a
rejected_answer field: the answer from the counterpart sample (same
edited/original image pair, opposite… See the full description on the dataset page: https://huggingface.co/datasets/vanloc1808/pico-banana-smolvlm-format-with-rejected-answer.hendrycks_math_with_answersstackexchange-question-answering
SYNTHETIC-1
This is a subset of the task data used to construct SYNTHETIC-1. You can find the full collection here
imo-answerbench
IMO-AnswerBench
Dataset Description
IMO-AnswerBench is a benchmark dataset for evaluating the mathematical reasoning capabilities of large language models. It consists of 400 challenging short-answer problems from the International Mathematical Olympiad (IMO) and other sources.
This dataset is part of the IMO-Bench suite, released by Google DeepMind in conjunction with their 2025 IMO gold medal achievement.
Supported Tasks and Leaderboards
The primary task… See the full description on the dataset page: https://huggingface.co/datasets/Hwilner/imo-answerbench.enwiki
English Wikipedia as clean md
This dataset is a cleaned, structurally faithful approximation of the English Wikipedia article corpus in Answer.AI's canonical md dialect. It was produced from the Wikimedia dump dated 20260901 by Answer.AI's wiki2dataset pipeline.
It is designed for language-model training and for agent/RAG systems. The articles configuration provides complete documents for continued pretraining, corpus analysis, rechunking, and task-specific dataset creation. The… See the full description on the dataset page: https://huggingface.co/datasets/answerdotai/enwiki.iheval-benign-answerssycophancy_eval_answersimplewiki
Simple English Wikipedia as clean md
This dataset is a cleaned, structurally faithful approximation of the Simple English Wikipedia article corpus in Answer.AI's canonical md dialect. It was produced from the Wikimedia dump dated 20260901 by Answer.AI's wiki2dataset pipeline.
It is designed for language-model training and for agent/RAG systems. The articles configuration provides complete documents for continued pretraining, corpus analysis, rechunking, and task-specific dataset… See the full description on the dataset page: https://huggingface.co/datasets/answerdotai/simplewiki.multi-context-long-answer-datasetmodel_answer
Dataset Card for "livebench/model_answer"
LiveBench is a benchmark for LLMs designed with test set contamination and objective evaluation in mind. It has the following properties:
LiveBench is designed to limit potential contamination by releasing new questions monthly, as well as having questions based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses.
Each question has verifiable, objective ground-truth answers, allowing hard questions to be… See the full description on the dataset page: https://huggingface.co/datasets/livebench/model_answer.msmarco_answerai_colbert_small_embeddings
MS MARCO ColBERT Embeddings
Pre-computed ColBERT embeddings for MS MARCO using PyLate and answerdotai/answerai-colbert-small-v1.
Dataset Structure
The dataset contains:
data/corpus/: 177 parquet files with document embeddings
data/queries/: 11 parquet files with query embeddings
data/qrels/train.parquet: Relevance judgments (532,751 pairs)
Usage
from datasets import load_dataset
# Load from directory (recommended for large datasets)
corpus =… See the full description on the dataset page: https://huggingface.co/datasets/WenxingZhu/msmarco_answerai_colbert_small_embeddings.yahoo-answers
Dataset Card for Yahoo Answers
This dataset is a collection of pairs containing titles, questions, and answers collected from Yahoo Answers. See the Yahoo Answers dataset for additional information. This dataset can be used directly with Sentence Transformers to train embedding models.
Dataset Subsets
title-question-answer-pair subset
Columns: "question", "answer"
Column types: str, str
Examples:{
'question': "why doesn't an optical mouse work on a glass… See the full description on the dataset page: https://huggingface.co/datasets/sentence-transformers/yahoo-answers.IMO-AnswerBench
IMO-AnswerBench
Dataset Description
IMO-AnswerBench is a benchmark dataset for evaluating the mathematical reasoning capabilities of large language models. It consists of 400 challenging short-answer problems from the International Mathematical Olympiad (IMO) and other sources.
This dataset is part of the IMO-Bench suite, released by Google DeepMind in conjunction with their 2025 IMO gold medal achievement.
Supported Tasks and Leaderboards
The primary task… See the full description on the dataset page: https://huggingface.co/datasets/OpenEvals/IMO-AnswerBench.task453_swag_answer_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task453_swag_answer_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks}… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task453_swag_answer_generation.answers-with-reasoning-mmlu-pro
answers-with-reasoning-mmlu-pro
Self-distillation SFT corpus: Qwen3-8B-Instruct's own correct
chain-of-thought rollouts on MMLU-Pro multiple-choice questions
(general-QA domain).
Generation
Source problems: TIGER-Lab/MMLU-Pro test split (12,032 multiple-choice questions across 14 subject categories).
Sampling model: qwen/qwen3-8b via OpenRouter (providers: Alibaba, AtlasCloud) with reasoning enabled.
Sampling parameters: temperature=0.6, top_p=0.95, max_tokens=8000.… See the full description on the dataset page: https://huggingface.co/datasets/abhayesian/answers-with-reasoning-mmlu-pro.medical-question-answering-splittask067_abductivenli_answer_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task067_abductivenli_answer_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task067_abductivenli_answer_generation.yahoo_answers_topicsanswerable_tydiqa
Dataset Card for "answerable-tydiqa"
Dataset Summary
TyDi QA is a question answering dataset covering 11 typologically diverse languages.
Answerable TyDi QA is an extension of the GoldP subtask of the original TyDi QA dataset to also include unanswertable questions.
Dataset Structure
The dataset contains a train and a validation set, with 116067 and 13325 examples, respectively. Access them with
from datasets import load_dataset
dataset =… See the full description on the dataset page: https://huggingface.co/datasets/copenlu/answerable_tydiqa.nq-question-answeronly