datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
met-office-uk-deterministic-solar
Met Office UK Deterministic Dataset (Zarr Format)
Description
This dataset is a subset of the Met Office UK Deterministic Dataset, converted from the original NetCDF format into Zarr format for modern data analysis. The Zarr files are packaged as .zarr.zip archives for efficient storage and transfer.
The subset focuses on specific variables and configurations, which are detailed below. Researchers and developers can use this subset for applications in climate science… See the full description on the dataset page: https://huggingface.co/datasets/openclimatefix/met-office-uk-deterministic-solar.officeqa
OfficeQA
Dataset Summary
OfficeQA is a grounded reasoning benchmark by Databricks for evaluating model and agent performance on end-to-end reasoning over real-world documents.
The benchmark consists of question–answer pairs that require reasoning over historical U.S. Treasury Bulletin documents (1939–2025), which contain dense financial tables, charts, and narrative text. OfficeQA is designed to test retrieval, tool use, and multi-step reasoning in… See the full description on the dataset page: https://huggingface.co/datasets/databricks/officeqa.officeqa
OfficeQA manifest (nearai-bench packaging)
Harness-ready question manifest for
databricks/officeqa — document-grounded
QA over U.S. Treasury Bulletins (1939–2025). 246 items in full,
8 in smoke (a 4-easy/4-hard subset for pipeline checks).
from datasets import load_dataset
ds = load_dataset("NEAR-AI/officeqa", split="full")
⚠️ This is the manifest only — documents are NOT included
Unlike our pinchbench and
clawbench exports, the
source corpus is not bundled here.… See the full description on the dataset page: https://huggingface.co/datasets/NEAR-AI/officeqa.OmegaUse-OfficeVal
OmegaUse-OfficeVal
Benchmarking LLM Agents on Long-Horizon Office-Suite Tasks with Economic Grounding
OmegaUse-OfficeVal is a benchmark for evaluating LLM agents on long-horizon,
real-world office-suite tasks that span word-processing documents, spreadsheets,
presentations, and cross-file productivity workflows. Tasks are derived from
authentic office requests proposed by practitioners and drawn from freelance
platforms, grounding the benchmark in real economic demand. Each task… See the full description on the dataset page: https://huggingface.co/datasets/baidu-frontier-research/OmegaUse-OfficeVal.officeqa-pro-v2
OfficeQA Pro v2
Dataset Summary
OfficeQA Pro v2 is a grounded reasoning benchmark by Databricks for evaluating model and agent performance on end-to-end reasoning over real-world documents.
The benchmark consists of question–answer pairs that require reasoning over two centuries of U.S. Federal Accounts of Receipts and Expenditures reporting (1793–2024) — Combined Statements of Receipts, Outlays, and Balances of the United States Government, together with earlier… See the full description on the dataset page: https://huggingface.co/datasets/databricks/officeqa-pro-v2.Office-Home
Dataset Card for Office-Home
This is a FiftyOne dataset with 15588 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
import fiftyone.utils.huggingface as fouh
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub("Voxel51/Office-Home")
# Launch the App
session = fo.launch_app(dataset)
Dataset Details… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/Office-Home.osworld_office_RLoffice-home
Dataset Card for Office-Home
The Office-Home dataset has been created to evaluate domain adaptation algorithms for object recognition using deep learning. It consists of images from 4 different domains: Artistic images, Clip Art, Product images and Real-World images. For each domain, the dataset contains images of 65 object categories found typically in Office and Home settings.
Dataset Details
The dataset information is based on the original dataset website:… See the full description on the dataset page: https://huggingface.co/datasets/flwrlabs/office-home.officeqa-checkpoint-eval-data
Checkpoint evaluation plot data
Snapshot: 2026-09-14T16:26:45.684890+00:00. Aggregate inputs to notes/Sept-2-2026.md performance figures.
No model execution, grading, publication, or source-result changes were performed to make this export.
Contents
checkpoint_evaluations: 454 checkpoint rows, one evaluation per run/iteration/protocol; score, mean output tokens, mean steps, and the existing two-sided 95% confidence bounds.
pareto_points: current mean-token/USD… See the full description on the dataset page: https://huggingface.co/datasets/YWZBrandon/officeqa-checkpoint-eval-data.taste-rob-office-3044-to-6052
TASTE-Rob Office_3044_to_6052 Video Dataset
This dataset contains 185 videos from the Office_3044_to_6052 scene of TASTE-Rob dataset.
Usage
from huggingface_hub import hf_hub_download
# Download a specific video
video_path = hf_hub_download(
repo_id="charlychan123/taste-rob-office-3044-to-6052",
filename="video_name.mp4",
repo_type="dataset"
)
Citation
If you use this dataset, please cite the original TASTE-Rob paper.
OfficeComprehensionBenchmark
OfficeComprehensionBenchmark (OCB)
OCB is a benchmark for evaluating document comprehension and grounded reasoning over Microsoft Office files (Word, Excel, PowerPoint). It comprises two tracks:
File Fidelity Q&A — measures structural and visual perception of document artifacts (text, tables, charts, formulas, formatting, embedded objects).
Domain Q&A — measures expert-level reasoning over real-world business documents across 12 industries.
Companion repository… See the full description on the dataset page: https://huggingface.co/datasets/microsoft/OfficeComprehensionBenchmark.OfficeBench_excel_part数据集的Task大部分是分析excel文件输出为其他形式的文件，例如分析excel之后结果输出到word中。
纯涉及到的excel内部进行操作的数据有点少，于是涉及到分析excel文件的数据均保留，如果有需要纯excel操作的数据我后续进行筛选
taste-rob-office-6053-to-9429
TASTE-Rob Office_6053_to_9429 Video Dataset
This dataset contains 251 videos from the Office_6053_to_9429 scene of TASTE-Rob dataset.
Usage
from huggingface_hub import hf_hub_download
# Download a specific video
video_path = hf_hub_download(
repo_id="charlychan123/taste-rob-office-6053-to-9429",
filename="video_name.mp4",
repo_type="dataset"
)
Citation
If you use this dataset, please cite the original TASTE-Rob paper.
met-office-uk-deterministic-zarr
Met Office UK Deterministic Dataset (Zarr Format)
Description
This dataset is a subset of the Met Office UK Deterministic Dataset, converted from the original NetCDF format into Zarr format for modern data analysis. The Zarr files are packaged as tar archives for efficient storage and transfer.
The subset focuses on specific variables and configurations, which are detailed in the met_office_uk_data_config.yaml file included in this repository. Researchers and developers… See the full description on the dataset page: https://huggingface.co/datasets/jcamier/met-office-uk-deterministic-zarr.open_door_officeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "deerbaby",
"total_episodes": 62,
"total_frames": 31000,
"total_tasks": 1,
"total_videos": 0,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 50,
"splits": {
"train": "0:62"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/silverlife/open_door_office.open_door_left_office_hdfoffice_multimodal_sweep
Office Multimodal Manipulation Sweep
A dual-arm (ALOHA) manipulation dataset collected in the RoboPRO / RoboTwin
simulator. Every task is run over many scene seeds and four distinct planner
motion modes, so the same (task, seed) is recorded four ways that differ only
in how the arm moves — a controlled source of trajectory-level multimodality.
Contents
20 office tasks × 20 scene seeds × 4 motion modes = 1,600 episodes.
Outcome tally: 919 mode_success · 646… See the full description on the dataset page: https://huggingface.co/datasets/choijoshua16/office_multimodal_sweep.office-rover-2task-raw
office-rover-2task-raw
LeRobot v2.1 dataset for a single SO-101 arm: "pick up the red / green cube
and put it in the box" with both cubes on the table. Two language-conditioned
tasks (instructions in Russian). Built for fine-tuning NVIDIA Isaac GR00T N1.7
on one 24 GB RTX 4090 — recipe, patches and 1800 evaluated attempts:
https://github.com/VShirokun/gr00t-on-4090
What is in it: the raw, un-engineered demonstrations: cubes in fixed orientation, 240×320 cameras. Trained as-is… See the full description on the dataset page: https://huggingface.co/datasets/VShirokun/office-rover-2task-raw.office-rover-miss-v21
office-rover-miss-v21
LeRobot v2.1 dataset for a single SO-101 arm: "pick up the red / green cube
and put it in the box" with both cubes on the table. Two language-conditioned
tasks (instructions in Russian). Built for fine-tuning NVIDIA Isaac GR00T N1.7
on one 24 GB RTX 4090 — recipe, patches and 1800 evaluated attempts:
https://github.com/VShirokun/gr00t-on-4090
What is in it: random cube yaw (0–90°) and 30 % deliberately failed grasps followed by scripted recovery — the data… See the full description on the dataset page: https://huggingface.co/datasets/VShirokun/office-rover-miss-v21.office-flight-tracker-london-tiles
London Streets.gl tiles
Generated 2026-09-17T12:29:11.517Z from © OpenStreetMap contributors.
Format: uncompressed Mapbox Vector Tiles generated by the
StrandedKitty/planetiler StreetsProfile, zooms 9–16.
OfficeHomeDatasetSimia-OfficeBench-SFT-30k
🐒 Simia-OfficeBench-SFT-30k:
Simia-OfficeBench-SFT-30k is the fully synthetic tool-agent dataset, designed to advance tool use for OfficeBench (2-apps and 3-apps). It comprises nearly 30k synthesized trajectories from 1-app task trajectories. Models fine-tuned on this dataset outperform much larger closed-source counterparts on OfficeBench (2-apps and 3-apps).
📄 Technical Report - Discover the methodology and technical details behind this dataset
💾 Github Repo - Access the… See the full description on the dataset page: https://huggingface.co/datasets/Simia-Agent/Simia-OfficeBench-SFT-30k.office_vortex_realsenseofficebench-automation
officebench-automation
OfficeBench: Benchmarking Language Agents across Multiple Applications for Office Automation
Dataset Description
This dataset is a standardized version of the original benchmark, prepared for easy evaluation of LLMs on planning tasks.
Splits
test: 300 samples
Features
{
"id": "Value(dtype='string', id=None)",
"task_id": "Value(dtype='string', id=None)",
"subtask_id": "Value(dtype='string', id=None)",
"num_apps":… See the full description on the dataset page: https://huggingface.co/datasets/tuandunghcmut/officebench-automation.so101_office_test_20250802This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "so101",
"total_episodes": 5,
"total_frames": 2240,
"total_tasks": 1,
"total_videos": 10,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:5"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/ktkd/so101_office_test_20250802.taste-rob-office-1-to-3043
TASTE-Rob Office_1_to_3043 Video Dataset
This dataset contains 449 videos from the Office_1_to_3043 scene of TASTE-Rob dataset.
Usage
from huggingface_hub import hf_hub_download
# Download a specific video
video_path = hf_hub_download(
repo_id="charlychan123/taste-rob-office-1-to-3043",
filename="video_name.mp4",
repo_type="dataset"
)
Citation
If you use this dataset, please cite the original TASTE-Rob paper.
so100_officeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.0",
"robot_type": "so100",
"total_episodes": 50,
"total_frames": 29880,
"total_tasks": 1,
"total_videos": 100,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 30,
"splits": {
"train": "0:50"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/vladfatu/so100_office.office-rover-grasp-hd-v21
office-rover-grasp-hd-v21
LeRobot v2.1 dataset for a single SO-101 arm: "pick up the red / green cube
and put it in the box" with both cubes on the table. Two language-conditioned
tasks (instructions in Russian). Built for fine-tuning NVIDIA Isaac GR00T N1.7
on one 24 GB RTX 4090 — recipe, patches and 1800 evaluated attempts:
https://github.com/VShirokun/gr00t-on-4090
What is in it: grasp-centred demonstrations at 480×640 without cube rotation — the base skill. The other half of… See the full description on the dataset page: https://huggingface.co/datasets/VShirokun/office-rover-grasp-hd-v21.DICE-BENCH
🎲 DICE-BENCH: Evaluating the Tool-Use Capabilities of Large Language Models in Multi-Round, Multi-Party Dialogues
🔗 Links for Reference
Repository: https://github.com/snuhcc/DICE-Bench
Paper: https://arxiv.org/abs/2506.22853
Project page: https://snuhcc.github.io/DICE-Bench/
Point of Contact: kyochul@snu.ac.kr
📖 Paper Description
DICE-BENCH is a benchmark that tests how well large language models can call external functions in realistic… See the full description on the dataset page: https://huggingface.co/datasets/OfficerChul/DICE-BENCH.titer-edgar-officers
titer · EDGAR officer corpus
4,206,080 attested person–company–role–date tuples from SEC Forms 3/4/5,
published as pointers rather than records, alongside the frozen
pre-registrations that were hash-published before any measurement ran.
edgar_officers.parquet: 4.2M rows, 230,405 distinct people, 20,266 issuers,
2006q1–2026q2.
Column
Meaning
accession
SEC accession number, the pointer that reconstructs the row
person_cik
Reporting-owner CIK. Never recycled by the… See the full description on the dataset page: https://huggingface.co/datasets/caiotheodoro/titer-edgar-officers.