datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
scaling-data-constrained-llms
Scaling Data-Constrained Language Models with Synthetic Data
This repository provides the pre-training corpora used in Scaling Data-Constrained Language Models with Synthetic Data (Findings of EACL 2026).
Overview
This repository contains multiple corpora designed to study data augmentation strategies for pre-training Japanese LLMs under a data-constrained data setting.
Starting from a limited Japanese Web corpus and a larger English Web corpus, we construct three… See the full description on the dataset page: https://huggingface.co/datasets/llm-jp/scaling-data-constrained-llms.DEBATE
DEBATE: Diverse Multi-Agent Debates
This dataset is presented in the paper "MALLM: Multi-Agent Large Language Models Framework".
Citation
comming soon.
juris-tcuJurisTCU is a Brazilian Portuguese legal IR resource built from the curated jurisprudence collection of the Brazilian Federal Court of Accounts (TCU), from which we derive the benchmark subset used here. In its original form, the dataset contains 16,045 jurisprudence documents organized into more than 20 fields (metadata and textual fields). The most relevant are ENUNCIADO and EXCERTO, which correspond, respectively, to a summary of the ruling and to the excerpt from the decision that supports… See the full description on the dataset page: https://huggingface.co/datasets/ufca-llms/juris-tcu.BR-TaxQAThis dataset corresponds to BR-TaxQA-R, a collection derived from materials of the Brazilian Federal Revenue Service (Receita Federal) on personal income tax (IRPF). It contains 715 questions with their corresponding answers and introduces user-oriented, FAQ-style query formulations in a legal-tax domain.
Some questions are explicitly linked to other related questions. In our relevance design, the immediate answer to the queried question is treated as the primary positive with score = 2, while… See the full description on the dataset page: https://huggingface.co/datasets/ufca-llms/BR-TaxQA.normas-tcuNormasTCU is a Brazilian Portuguese legal IR test collection composed of normative documents from the Brazilian Federal Court of Accounts (TCU). These normative acts may have internal effects (e.g., rules governing internal procedures) or external effects (e.g., rules regulating how the court interacts with other public institutions) and differ from jurisprudential documents in both purpose and structure. Jurisprudential documents typically describe specific cases and present the legal… See the full description on the dataset page: https://huggingface.co/datasets/ufca-llms/normas-tcu.juaJUÁ-Juris is centered on jurisprudence drawn from the curated jurisprudence collection of the Brazilian Federal Court of Accounts (TCU). In this collection, each instance contains an enunciado and an excerto: the enunciado is an abstractive summary of the ruling, while the excerto is the passage from the ruling that supports that summary. In our retrieval setup, the enunciado serves as the query, and the corresponding excerto is treated as the ground-truth positive passage. Within the… See the full description on the dataset page: https://huggingface.co/datasets/ufca-llms/jua.llmsql-benchmark
LLMSQL Benchmark
⚠️ A newer version of this dataset is available:👉 https://huggingface.co/datasets/llmsql-bench/llmsql-2.0
This benchmark is designed to evaluate text-to-SQL models. For usage of this benchmark see https://github.com/LLMSQL/llmsql-benchmark.
Arxiv Article: https://arxiv.org/abs/2510.02350
Files
tables.jsonl — Database table metadata
questions.jsonl — All available questions
train_questions.jsonl, val_questions.jsonl, test_questions.jsonl — Data… See the full description on the dataset page: https://huggingface.co/datasets/llmsql-bench/llmsql-benchmark.llmsql-2.0
LLMSQL 2.0: a small, hard and verified Text-to-SQL benchmark
LLMSQL 2.0 is a test-only benchmark of 2,000 natural-language questions over 942 real Wikipedia tables.
Each question comes with a reference SQLite query and its verified answer. The benchmark is designed to be:
Small. 2,000 questions instead of the 15.8k test questions of the previous LLMSQL release, so a full run of a large
model takes minutes, not hours.
Hard. gpt-oss-120b, which solves about 93% of the earlier… See the full description on the dataset page: https://huggingface.co/datasets/llmsql-bench/llmsql-2.0.Multi-turn_Long-context_Benchmark_for_LLMs
LoopServe: An Adaptive Dual-phase LLM Inference Acceleration System for Multi-Turn Dialogues
Arxiv: https://www.arxiv.org/abs/2507.13681
Huggingface: https://huggingface.co/papers/2507.13681
Introduction
LoopServe Multi-Turn Dialogue Benchmark is a comprehensive evaluation dataset comprising multiple diverse datasets designed to assess large language model performance in realistic conversational scenarios.
Unlike traditional benchmarks that place queries only at the end… See the full description on the dataset page: https://huggingface.co/datasets/TreeAILab/Multi-turn_Long-context_Benchmark_for_LLMs.LLM-self-identification Self Identification – Give your Language model an identity
About Self Identification
Self identification is training set SupraLabs curated for developers/trainers to experiment with to let your language model know about their identity.
Self identification let your LM know these information about them:
Model ID
Model Name
Model Description
Model Creator
Model Family
Model Architecture
Parameter Count
Knowledge Cutoff
Here is an example from the dataset:
If… See the full description on the dataset page: https://huggingface.co/datasets/SupraLabs/LLM-self-identification.llm-system-prompts-benchmark
Dataset Card for Dataset Name
This datset is a collection of 100 system prompts for large language models.
Dataset Details
Dataset Description
These 100 system prompts test a model's ability to follow grammatical patterns; answer basic multiple choice questions; act according to a particular persona; memorize information; and speak in French.
Files:
hundred_system_prompts.py: refer to this to see the (prompt, probe, function) triplets, as well as the… See the full description on the dataset page: https://huggingface.co/datasets/Naomibas/llm-system-prompts-benchmark.llm-sfm-safety-eval
LLM x SFM Safety Evaluation
When a general-purpose language model interprets the output of a specialist
science foundation model (a protein, genomic, RNA, or chemistry model), does its
safety behavior recognize the scientific content, or only the surface form of the
request?
This repository is the empirical core of a study of that question: the evaluation
harness, the redacted aggregate results, and the measurement specifications behind
four findings about how deployed Claude… See the full description on the dataset page: https://huggingface.co/datasets/jang1563/llm-sfm-safety-eval.LLM_SMARTHOME2Korean-YouTube-Comment-Sentiment-Dataset
Korean YouTube Comment Sentiment Dataset
Data Overview
Summary
본 데이터셋은 유튜브에서 수집된 한국어 댓글 5,482개와 이에 대응하는 감정 레이블(긍정, 부정, 중립, 불명확)로 구성된 감정 분류용 데이터셋입니다.
주요 레이블: 긍정, 부정, 중립, 불명확
Features
수집 대상: 요리, 뷰티, 게임, 여행, 쇼핑 등 분야의 10만 명 이상 구독자를 보유한 유튜브 채널
형식: JSON (id, text, label)
검수: 한국인 검수자에 의한 수작업 라벨링 및 교차 검토
본 데이터셋은 구어체, 이모지, 줄임말 등 실제 사용자 표현이 반영되어 있습니다.
Dataset Structure
Dataset Fields
Field
Type
Description
id
string
각 댓글의… See the full description on the dataset page: https://huggingface.co/datasets/LLM-SocialMedia/Korean-YouTube-Comment-Sentiment-Dataset.LLM-self-identification
LLM Identity · Give your LLM an identity
Self Identification
The Self-Identification Dataset, curated by Qyrou, is a specialized training resource designed to help developers and trainers establish clear self-identity awareness within language models. By incorporating this dataset, models can accurately learn and convey essential metadata about themselves, including their Model ID, Model Name, Model Description, Model Creator, Model Family, Model Architecture, Parameter Count… See the full description on the dataset page: https://huggingface.co/datasets/mithulaartigala/LLM-self-identification.Electricity_Load_Forecasting_using_LLMSvalkompass-2026-llms
Valkompass 2026 × LLMs
How do 50 popular large language models answer the 35
questions in SVT's Swedish election compass (Valkompass 2026, Riksdag) — and which
of the 8 Riksdag parties does each model end up closest to?
Unlike comparisons that query chat products (ChatGPT, Gemini, Claude, Grok web UIs),
which have web-search / tools and act as agents, this dataset probes the raw model
weights only, via the OpenRouter API, with no system prompt, no tools, no web access.… See the full description on the dataset page: https://huggingface.co/datasets/nordan-ai/valkompass-2026-llms.llm-serving-bench-data
llm-serving-bench data
Derived data for the LLM serving capacity study in https://github.com/hoangphu7122002/llm-serving-bench,
built by src/workloads/corpus.py (sources, versions and licenses: src/workloads/README.md).
Each file keeps the license of its source:
corpus/chat.jsonl — user turns sampled from ShareGPT V3 (anon8231489123/ShareGPT_Vicuna_unfiltered,
Apache-2.0 on the Hub; conversations with ChatGPT — research use).
corpus/rag.jsonl — English Wikipedia 20231101… See the full description on the dataset page: https://huggingface.co/datasets/hoangphu7122002ai/llm-serving-bench-data.llm-skills-registry
LLM Skills Registry
226 first-party skills and capabilities across 13 AI assistant ecosystems, each with a category, a description, a usage example, and a link to the vendor's own documentation.
This is the data behind SkillsAllYouNeed.
What counts as a skill here
A skill is a capability the vendor itself ships: something the assistant can do, triggered in a documented way. Third-party plugins and community prompt packs are out of scope, which is what keeps every… See the full description on the dataset page: https://huggingface.co/datasets/kishormorol/llm-skills-registry.self-distillation-LLMsGitHub
L-MTP: Leap Multi-Token Prediction Beyond Adjacent Context for Large Language Models
An extendable MTP paradigm with leaping capability for both training and inference.
Large Language Models (LLMs) are typically trained and deployed using Next-Token Prediction (NTP), an inherently sequential process that limits contextual coverage and inference efficiency. To address this, we propose Leap Multi-Token Prediction (L-MTP), an innovative method that extends Multi-Token Prediction… See the full description on the dataset page: https://huggingface.co/datasets/xiao-hao/self-distillation-LLMs.repro-evaluating-llms-comparative-signals-traces
Agent traces
Agent sessions published from a Trackio Logbook.
Alpaca-ShareGPTllm-security-leaderboard-requeststts_llmsr_data
tts_llmsr_data
repro-codetaste-can-llms-generate-human-level-code-refactorings-traces
Agent traces
Agent sessions published from a Trackio Logbook.
LLM-Science-exam-datasetllm-seo-research
LLM SEO Research
Summary
Structured research-grade notes on LLM SEO, generative retrieval, citation behavior, and RAG-aligned page design. Intended to seed embedding indexes and synthetic QA without generic marketing fluff. Each row is a self-contained markdown document with abstract and topic tags.
Hub target: nebulatech/llm-seo-researchSplit: train (expand with validation when gold Q/A exists)
Terminology
AI SEO — Optimizing owned content and structured… See the full description on the dataset page: https://huggingface.co/datasets/nebulatech/llm-seo-research.quati
Quati 1M Jua-like
This dataset is a structural conversion of unicamp-dl/quati into a layout compatible with the repository organization used by ufca-llms/jua.
It uses the Quati 1M document collection and preserves the source evaluation setup instead of creating synthetic supervised training labels.
Dataset Summary
corpus.jsonl: 1,000,000 passages in JSONL format with fields _id, title, and text
queries.jsonl: 200 topics in JSONL format with fields _id and text… See the full description on the dataset page: https://huggingface.co/datasets/ufca-llms/quati.sud_resh_evaluated_llms_answers
📊 Результаты Оценки Больших Языковых Моделей на Бенчмарке Судебных Решений
В данном документе представлен анализ производительности 15 больших языковых моделей (LLM), протестированных на специализированном бенчмарке, который включает 105 000 записей из судебных решений России. Оценка проводилась по 10 различным категориям права (например, трудовое, уголовное, гражданское) и 7 типам инструкций (например, изложение исковых требований, анализ доказательств, итоговое решение).
Ответы… See the full description on the dataset page: https://huggingface.co/datasets/lawful-good-project/sud_resh_evaluated_llms_answers.OpenSFT-ita
OpenSFT ITA