datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
IntPhys2
IntPhys 2
Dataset |
Hugging Face |
Paper |
Blog
IntPhys 2 is a video benchmark designed to evaluate the intuitive physics understanding of deep learning models. Building on the original IntPhys benchmark, IntPhys 2 focuses on four core principles related to macroscopic objects: Permanence, Immutability, Spatio-Temporal Continuity, and Solidity. These conditions are inspired by research into intuitive physical understanding emerging during early childhood. IntPhys 2… See the full description on the dataset page: https://huggingface.co/datasets/facebook/IntPhys2.Multi-IF
Dataset Summary
We introduce Multi-IF, a new benchmark designed to assess LLMs' proficiency in following multi-turn and multilingual instructions. Multi-IF, which utilizes a hybrid framework combining LLM and human annotators, expands upon the IFEval by incorporating multi-turn sequences and translating the English prompts into another 7 languages, resulting in a dataset of 4501 multilingual conversations, where each has three turns. Our evaluation of 14 state-of-the-art LLMs on… See the full description on the dataset page: https://huggingface.co/datasets/facebook/Multi-IF.ExploreToM
Data sample for ExploreToM: Program-guided adversarial data generation for theory of mind reasoning
ExploreToM is the first framework to allow large-scale generation of diverse and challenging theory of mind data for robust training and evaluation.
Our approach leverages an A* search over a custom domain-specific language to produce complex story structures and novel, diverse, yet plausible scenarios to stress test the limits of LLMs.
Our A* search procedure aims to find… See the full description on the dataset page: https://huggingface.co/datasets/facebook/ExploreToM.AdvancedIF
Dataset Summary
We introduce AdvancedIF, a new benchmark featuring over 1,600 prompts and expert-curated rubric designed to assess LLMs' proficiency in
Complex instruction following: each prompt has 6+ instructions with combination
of one, format, style, structure, length, negative constraints, spelling, and inter-conditional instructions;
Multi-turn instruction following: the ability to follow instruction carried from previous;
System prompt steerability: The ability to follow… See the full description on the dataset page: https://huggingface.co/datasets/facebook/AdvancedIF.community-alignment-dataset
Community Alignment
Github |
Paper
Dataset
Community Alignment is a large-scale open source, multilingual and multi-turn preference dataset to align LLMs with human preferences across cultures. Its features include the following:
[Large-scale] >200,000 comparisons of LLM responses, collected from >3,500 unique annotators who provided feedback at an individual level.
[Multilingual] Contains comparisons in English, French, Italian, Hindi, and Portuguese. 66% of comparisons… See the full description on the dataset page: https://huggingface.co/datasets/facebook/community-alignment-dataset.EgoAVU_data
[CVPR2026 HIGHLIGHT] EgoAVU, [ICASSP2026 Oral] Exploring Audio Hallucination in Egocentric Video Understanding
Official Implementation of EgoAVU: Egocentric Audio-Visual Understanding and Exploring Audio Hallucination in Egocentric Video Understanding
See our github for the code and setup instructions.
Check out our homepage, paper (CVPR) and paper (ICASSP) for more information.
We introduce EgoAVU, a scalable and automated data engine to enable egocentric audio–visual… See the full description on the dataset page: https://huggingface.co/datasets/facebook/EgoAVU_data.OMC25
Open Molecular Crystals 2025 (OMC25)
Dataset
Dataset
LICENSE: The OMC25 dataset is provided under a CC-BY-4.0 license
OMC25 represents the largest high quality molecular crystal DFT dataset.
OMC25 was generated at the PBE-D3 level of theory as implemented in Vienna Ab initio Simulation Package (VASP). OMC25 includes structures sampled from relaxation trajectories of molecular crystals generated by Genarris 3.0 starting from molecules in the OE62 dataset.… See the full description on the dataset page: https://huggingface.co/datasets/facebook/OMC25.CIMemories
CIMemories: A Compositional Benchmark for Contextual Integrity of Persistent Memory in LLMs
Paper
Large Language Models (LLMs) increasingly use persistent memory from past interactions to enhance personalization and task performance. However, this memory introduces critical risks when sensitive information is revealed in inappropriate contexts. We present CIMemories, a benchmark for evaluating whether LLMs appropriately control information flow from memory based on task context.… See the full description on the dataset page: https://huggingface.co/datasets/facebook/CIMemories.airs-bench
AIRS-Bench: a Suite of Tasks for Frontier AI Research Science Agents
The AI Research Science Benchmark (AIRS-Bench) quantifies the autonomous research abilities of LLM agents in the area of machine learning. AIRS-Bench comprises 20 tasks from state-of-the-art machine learning papers spanning diverse domains: NLP, Code, Math, biochemical modelling, and time series forecasting.
Each task is specified by a ⟨problem, dataset, metric⟩ triplet and a SOTA value. The agent receives the… See the full description on the dataset page: https://huggingface.co/datasets/facebook/airs-bench.HoneyBee
HoneyBee: Data Recipes for Vision-Language Reasoners
This is the official data release for the paper: https://arxiv.org/abs/2510.12225.
Github Repo: https://github.com/facebookresearch/HoneyBee_VLM.
Abstract
Recent advances in vision-language models (VLMs) have made them highly effective at reasoning tasks. However, the principles underlying the construction of performant VL reasoning training datasets remain poorly understood. In this work, we introduce several data… See the full description on the dataset page: https://huggingface.co/datasets/facebook/HoneyBee.content_rephrasing
Message Content Rephrasing Dataset
Introduced by Einolghozati et al. in Sound Natural: Content Rephrasing in Dialog Systems https://aclanthology.org/2020.emnlp-main.414/
We introduce a new task of rephrasing for amore natural virtual assistant. Currently, vir-tual assistants work in the paradigm of intent-slot tagging and the slot values are directlypassed as-is to the execution engine. However,this setup fails in some scenarios such as mes-saging when the query given by the user… See the full description on the dataset page: https://huggingface.co/datasets/facebook/content_rephrasing.facebook-personality-recognition-wcpr13The Workshop on Computational Personality Recognition 2013 was a competition based on this Facebook dataset.
The purpose is to predict the personality scores or classes from text and ego-network data
reference paper: https://ojs.aaai.org/index.php/ICWSM/article/view/14467/14316
toolverifier
TOOLVERIFIER: Generalization to New Tools via Self-Verification
This repository contains the ToolSelect dataset which was used to fine-tune Llama-2 70B for tool selection.
Data
ToolSelect data is synthetic training data generated for tool selection task using Llama-2 70B and Llama-2-Chat-70B.
It consists of 555 samples corresponding to 173 tools.
Each training sample is composed of a user instruction, a candidate set of tools that includes the
ground truth tool, and a… See the full description on the dataset page: https://huggingface.co/datasets/facebook/toolverifier.Y-NQ
Dataset Card for Y-NQ
The dataset is available in this csv file. The dataset is licensed under the Apache 2.0 license
Dataset Description
Question ID: Unique identifier from Natural Question
Split: Training or validation split from Natural Question
English Document: English text document
English Question: Question in English
English Long Answer: Detailed answer in English
English Short Answer: Brief answer in English
Yorùbá Document: Yorùbá text document
Yorùbá… See the full description on the dataset page: https://huggingface.co/datasets/facebook/Y-NQ.facebook_spam_detection
Facebook Spam Detection Dataset
Dataset Summary
This dataset contains 600 Facebook profiles with behavioral and activity features designed for spam detection in social media. The dataset enables binary classification to distinguish between spam accounts (Label=1) and legitimate accounts (Label=0), providing insights into spammer behavior patterns on Facebook.
Dataset Details
Total Samples: 600 profiles
Classes: Binary (0 = Legitimate, 1 = Spam)
Class… See the full description on the dataset page: https://huggingface.co/datasets/nahiar/facebook_spam_detection.Facebook_Sinhala_Hate_Speechbeyond_the_lab_neurips_paperSCRuB-dataset
SCRuB — Social Concept Reasoning under Rubric-Based Evaluation
SCRuB is a dataset suite for studying how large language models handle socially sensitive, open-ended essay prompts. It comprises three components:
Component
Description
Rows
SCRuBSample
30 curated study prompts used as stimuli in a human annotation study
30
SCRuBAnnotations
Expert essays, model responses, and quality judgments from a two-task annotation study
300 + 78 + 20 + 900 + 900
SCRuBEval4,711… See the full description on the dataset page: https://huggingface.co/datasets/facebook/SCRuB-dataset.FacebookDecadeCorporaFacebook-dataset10k-facebook-chatsfacebook-users-dataFacebook Users Engagement Analysis
Author: Tamara Banaim
Dataset: Pseudo Facebook Dataset (Kaggle, uploaded to Hugging Face)
Overview-
This project analyzes data from 99,003 Facebook users, focusing on demographic information and engagement metrics such as likes given, likes received, friend count, and account tenure. The analysis explores how age and user activity are related, and what factors influence engagement on the platform.
Objective-
To examine how age and… See the full description on the dataset page: https://huggingface.co/datasets/tamarabanaim/facebook-users-data.