datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
pulse-sofroniew-emotion-concept-texts
Pulse Geometry: Sofroniew-Style Implicit Emotion Corpus
A contrastive corpus of 8,550 short stories (171 emotions × 50 topics) that convey
a target emotion implicitly — through behavior, sensation, dialogue, internal
thought, or environmental description, but never by naming the emotion. Each story
is scored on a four-axis rubric by Claude Sonnet.
The corpus was built as the substrate for a geometry replication: probing whether
an emotion-vector layout analogous to Sofroniew et al.… See the full description on the dataset page: https://huggingface.co/datasets/jmccardle/pulse-sofroniew-emotion-concept-texts.ConceptCaps
Dataset Card for ConceptCaps
Dataset Summary
ConceptCaps is a music captioning dataset derived from MusicCaps, specifically designed for concept-based interpretability research in text-to-audio (TTA) generation systems. The dataset provides categorized musical concept annotations from a distilled taxonomy (200 unique tags) alongside natural language captions, enabling fine-grained analysis of how TTA models represent and generate musical concepts.
Unlike existing datasets… See the full description on the dataset page: https://huggingface.co/datasets/bsienkiewicz/ConceptCaps.ConceptVectors
ConceptVectors
🚀The first-ever parametric LLM Unlearning Benchmark!
We find current unlearning methods only modify model’s behavior without truly erasing encoded knowledge in parameters. For this, we present ConceptVectors Benchmark, with each vector strongly tied to a specific concept.
The ConceptVectors Benchmark for the paper "Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces".
Links
Paper: Intrinsic Test of Unlearning Using Parametric Knowledge… See the full description on the dataset page: https://huggingface.co/datasets/YihuaiHong/ConceptVectors.Statements-Of-Federal-Financial-Accounting-Concepts-And-Standards
Statements of Federal Financial Accounting Concepts and Standards
Dataset Summary
This dataset contains document-grounded question-and-answer samples based on the Statements of Federal Financial Accounting Concepts and Standards issued within the Federal accounting framework.
The source material establishes the concepts, principles, definitions, recognition criteria, measurement requirements, presentation standards, and disclosure expectations used in Federal… See the full description on the dataset page: https://huggingface.co/datasets/leeroy-jankins/Statements-Of-Federal-Financial-Accounting-Concepts-And-Standards.evolution-concept-development
Концепция развития. Нетривиальный взгляд на эволюцию / Concept of Development: A Non-Trivial Outlook on Evolution
Автор / Author: Владлен В.К. / Vladlen V.K.
Год / Year: 2022
Издательство / Publisher: Прометей (Москва)
ISBN: 978-5-00172-246-5
Лицензия / License: CC BY 4.0
RU — О датасете
Этот датасет содержит полный текст книги «Концепция развития. Нетривиальный взгляд на эволюцию» Владлена В.К., разбитый на 4 статьи. Книга излагает универсальный принцип… See the full description on the dataset page: https://huggingface.co/datasets/WladlenVK/evolution-concept-development.concept500-contrastiveData release from Concept DAS: Faithful Bi-Directional Model Steering via Distribution Matching and Distributed Interchange Interventions (ICLR 2026) (OpenReview).
Also used in Towards Steering without Sacrifice: Principled Training of Steering Vectors for Prompt-only Interventions.
Code: https://github.com/colored-dye/concept_das.
Contrastive training data built upon AxBench and Concept500.
Subsets:
2b_l10: 10th layer of Gemma2-2B.
2b_l20: 20th layer of Gemma2-2B.
9b_l20: 20th layer of… See the full description on the dataset page: https://huggingface.co/datasets/colored-dye/concept500-contrastive.conceptnet
ConceptNet 5.7
Dataset Description
Common-sense knowledge graph with everyday facts and relationships
Original Source: https://s3.amazonaws.com/conceptnet/downloads/2019/edges/conceptnet-assertions-5.7.0.csv.gz
Dataset Summary
This dataset contains RDF triples from ConceptNet 5.7 converted to HuggingFace dataset format
for easy use in machine learning pipelines.
Format: Originally csv, converted to HuggingFace Dataset
Size: 1.2 GB (extracted)
Entities: ~8M… See the full description on the dataset page: https://huggingface.co/datasets/CleverThis/conceptnet.deepscaler-teacher-sft-vllm-official-40k-clean-v4-conceptual
deepscaler-teacher-sft-vllm-official-40k-clean-v4-conceptual
Filtered version of ThunderstormXXL/deepscaler-teacher-sft-vllm-official-40k.
Filtering
reward filter enabled: False
minimum official reward: 1.0
scoring errors rejected: False
maximum text tokens: 32768
maximum response chars: 200000
near-duplicate SimHash hamming threshold: 4
required <think>...</think> and final boxed answer after reasoning
exact text/problem/response dedupe and near problem dedupe… See the full description on the dataset page: https://huggingface.co/datasets/ThunderstormXXL/deepscaler-teacher-sft-vllm-official-40k-clean-v4-conceptual.concept-to-root-dictionary
🌿 Concept-to-Root Dictionary
A mapping of universal concepts to Arabic triliteral roots for semantic compression
📖 Overview
This dataset provides mappings between universal semantic concepts and Arabic triliteral roots, designed for use as a compression layer in Large Language Models.
What are Arabic Roots?
Arabic uses a root-and-pattern morphological system where most words derive from 3-letter roots:
Root
Core Meaning
Derived Words… See the full description on the dataset page: https://huggingface.co/datasets/ArniSal/concept-to-root-dictionary.task693_mmmlu_answer_generation_conceptual_physics
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task693_mmmlu_answer_generation_conceptual_physics
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… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task693_mmmlu_answer_generation_conceptual_physics.flas-concept-46k
FLAS-Concept-46k
A model-agnostic concept-steering corpus used to train FLAS (Flow-based Activation Steering). 46,472 unique concepts, ~2.64 M rows (concept × prompt × steered response).
🌐 Project page: https://flas-ai.github.io
📄 Paper: https://arxiv.org/abs/2605.05892
💻 Code: https://github.com/flas-ai/FLAS
What's in it
Each row is a (prompt, steered response, concept) triple:
Column
Description
input
instruction / prompt
output
response… See the full description on the dataset page: https://huggingface.co/datasets/flas-ai/flas-concept-46k.technical-concept-simplifier-dataset
Technical Concept Simplifier Dataset
Overview
The Technical Concept Simplifier Dataset is a curated instruction-tuning dataset designed to help Large Language Models (LLMs) explain complex technical concepts in a clear, beginner-friendly, and educational manner.
This dataset was developed as part of an AI model adaptation and fine-tuning project focused on improving the ability of language models to simplify advanced computer science, software engineering, cloud… See the full description on the dataset page: https://huggingface.co/datasets/ujjawalbansal/technical-concept-simplifier-dataset.concept-injection-results
Concept Injection Results
Experimental data and analysis logs from the research "Replicating Introspection on Injected Content in Open-Source Language Models"
Overview
This project enables researchers to inject concept vectors directly into a model's hidden layers during inference, allowing investigation of whether language models can detect and report on artificially induced "thoughts."
This dataset contains the raw responses and processed evaluations produced… See the full description on the dataset page: https://huggingface.co/datasets/vazirani/concept-injection-results.concept-to-root-dictionary
🌿 Concept-to-Root Dictionary
A mapping of universal concepts to Arabic triliteral roots for semantic compression
📖 Overview
This dataset provides mappings between universal semantic concepts and Arabic triliteral roots, designed for use as a compression layer in Large Language Models.
What are Arabic Roots?
Arabic uses a root-and-pattern morphological system where most words derive from 3-letter roots:
Root
Core Meaning
Derived Words… See the full description on the dataset page: https://huggingface.co/datasets/root-semantic-research/concept-to-root-dictionary.concept-bridge-benchmark
Concept Bridge Benchmark
A benchmark for evaluating LLM creative semantic reasoning — the ability to construct valid multi-hop chains between unrelated concepts using typed semantic relations.
What is Concept Bridge?
A chess-like word game. Given a START concept and a TARGET concept, build a chain of exactly N hops where each hop connects two common-noun concepts via a named semantic relation (e.g., contains, made-of, causes, resembles). An LLM arbiter judges each hop.… See the full description on the dataset page: https://huggingface.co/datasets/ravimeduri76/concept-bridge-benchmark.