OfficeAgent
Collection
2 items • Updated
Datasets:
You need to agree to share your Hugging Face information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
OfficeInstruct is released for non-commercial research only, under the CC BY-NC 4.0 license. The data were contributed by invited participants and de-identified before release. Access requests are reviewed manually. Please describe your intended use below.
Log in or Sign Up to review the conditions and access this dataset content.
OfficeInstruct
OfficeInstruct is a dataset of long-trajectory, multi-turn interactions of tool-augmented office agents across documents (Doc), spreadsheets (Sheet), and presentations (PPT). Trajectories were collected with the OfficeAgent framework and filtered and repaired with a graph-based evidence evaluation that removes redundant tool cycles, execution errors, and unsupported generations.
Access
This dataset is gated. Please fill in the request form above; requests are reviewed manually. By requesting access you agree to use the data for non-commercial research only, under the CC BY-NC 4.0 license.
Structure
train-data/: training trajectories for Doc, Sheet, and PPT tasks (JSONL).eval-data/: evaluation data (JSONL), together with copies of public benchmarks used in the paper (TableBench, SpreadsheetBench, TReB, MMTU). These benchmarks are redistributed for convenience only and remain subject to their original licenses.
Citation
If you use OfficeInstruct, please cite:
@inproceedings{xu2026officeinstruct,
title = {OfficeInstruct: A Dataset for Tool-Augmented Office Agents on Long-Trajectory Generative Tasks},
author = {Xu, Weikai and Jiang, Zhizheng and Jiang, Lushuo and Huang, Kun and Liu, Yuxuan and Gao, Pengzhi and Liu, Wei and Luan, Jian and Hu, Xiaolin and An, Bo},
booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track},
year = {2026}
}
- Downloads last month
- 140