datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
CUA-Gym
CUA-Gym
CUA-Gym is a collection of verifiable computer-use agent tasks for reinforcement learning with verifiable rewards (RLVR). Each task pairs a natural-language instruction with executable setup artifacts and a Python reward function that checks task completion programmatically. For details, see the paper CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents.
This release contains the full public CUA-Gym task set after the necessary data review.… See the full description on the dataset page: https://huggingface.co/datasets/xlangai/CUA-Gym.A11y-CUA
A11y-CUA Dataset
A11y-CUA is a multimodal desktop interaction dataset for accessibility-focused computer-use agent research. It contains real task trajectories recorded on Windows across two human user groups and two computer use agents (CUAs), each operating under standard and accessibility-specific conditions. Every session captures: timestamped keyboard and mouse events, browser interaction logs, accessibility trees, screen video, and system audio.
The dataset is… See the full description on the dataset page: https://huggingface.co/datasets/berkeley-hci/A11y-CUA.Reduced-A11y-CUA
Reduced A11y-CUA
A structured-data-only subset of the A11y-CUA dataset on HuggingFace.
What's different from the full dataset
The full A11y-CUA dataset includes screen recordings (screen.mp4), system audio (system_audio.wav), and microphone audio (mic_audio.wav) for every session. These files account for the large majority of the dataset's total size.
This reduced version strips all video and audio files. Every other file is identical and complete:… See the full description on the dataset page: https://huggingface.co/datasets/berkeley-hci/Reduced-A11y-CUA.cuad-parquet
CUAD (Contract Understanding Attic Dataset) - Parquet
Modernized version of theatticusproject/cuad in Parquet format.
Original dataset uses a deprecated loading script. This version is pre-processed and ready to use with modern datasets library.
Structure
Each row contains a single text field: context + "\n" + question
Rows: 20,910 QA pairs
Source: 510 legal contracts
Split: train (all data, use your own split logic)
Usage
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/whpthomas/cuad-parquet.A11y-CUA
A11y-CUA
Dataset Summary
A11y-CUA is a multimodal computer-use dataset of accessibility-relevant desktop task executions.
Each task folder contains:
structured desktop action logs (*.json)
accessibility tree snapshots (*_a11y_tree.json)
task metadata (metadata_*.json)
screen recording (screen.mp4)
system audio (system_audio.wav)
The dataset includes both human users and model-agent runs under multiple accessibility conditions.
Dataset Structure
Top-level… See the full description on the dataset page: https://huggingface.co/datasets/ananyagm/A11y-CUA.better-cuad
Better CUAD 📜
This repository preserves the Contract Understanding Atticus Dataset (CUAD) where the full text and annotations of all contracts in the dataset have been joined together into a single jsonl files to facilitate loading with the Hugging Face 🤗 datasets library.
Enjoy!
A11y-CUA
A11y-CUA Dataset
A11y-CUA is a multimodal desktop interaction dataset for accessibility-focused computer-use agent research. It contains real task trajectories recorded on Windows across two human user groups and two computer use agents (CUAs), each operating under standard and accessibility-specific conditions. Every session captures: timestamped keyboard and mouse events, browser interaction logs, accessibility trees, screen video, and system audio.
The dataset is… See the full description on the dataset page: https://huggingface.co/datasets/hamid735/A11y-CUA.Reduced-A11y-CUA
Reduced A11y-CUA
A structured-data-only subset of the A11y-CUA dataset on HuggingFace.
What's different from the full dataset
The full A11y-CUA dataset includes screen recordings (screen.mp4), system audio (system_audio.wav), and microphone audio (mic_audio.wav) for every session. These files account for the large majority of the dataset's total size.
This reduced version strips all video and audio files. Every other file is identical and complete:… See the full description on the dataset page: https://huggingface.co/datasets/Uriel12Agl/Reduced-A11y-CUA.cua-harm-recovery
CUA Harm Recovery Preference Dataset
This dataset contains human preference judgments for evaluating recovery plans in computer use agent (CUA) harm scenarios introduced in Human-Guided Harm Recovery for Computer Use Agents
Dataset Summary
The dataset contains 1,130 annotated plan pairs across 226 unique harm scenarios in computer use contexts. Each pair consists of two recovery plans (Plan A and Plan B) that were evaluated by human annotators to determine which plan… See the full description on the dataset page: https://huggingface.co/datasets/christykl/cua-harm-recovery.cuad_own