---
title: ACuRL
canonical_url: "https://www.modelscope.cn/datasets/osunlp/ACuRL"
md_url: "https://www.modelscope.cn/datasets/osunlp/ACuRL.md"
repository: osunlp/ACuRL
last_updated: 2026-06-08
license: other
storage_size: "400 KB"
downloads: 40
stars: 0
---

# ACuRL

> ACuRL - osunlp 在 ModelScope 开源的数据集。ACuRL Curriculum Tasks

osunlp/ACuRL 是 ModelScope 魔搭社区上的数据集，存储大小 400 KB，采用 other 许可。

- **Repository**: osunlp/ACuRL
- **License**: other
- **Storage size**: 400 KB
- **Downloads**: 40
- **Stars**: 0
- **Last updated**: 2026-06-08

Source: https://www.modelscope.cn/datasets/osunlp/ACuRL

---

# ACuRL Curriculum Tasks

[Paper](https://arxiv.org/abs/2602.10356) | [GitHub](https://github.com/OSU-NLP-Group/ACuRL) | [Models](https://huggingface.co/collections/osunlp/acurl)

This dataset contains curriculum tasks generated by **ACuRL**, an **A**utonomous **Cu**rriculum **R**einforcement **L**earning framework for continually adapting computer-use agents to target environments with zero human data.

The dataset includes two splits, `qwen3vl` and `uitars`, corresponding to two base agents: Qwen3-VL-8B-Instruct and UI-TARS-1.5-7B. In both splits, curriculum tasks are generated with GPT-5.

Each row is a natural-language task generated for an environment and curriculum iteration. Each split contains 4,608 tasks in total. For each split, every data source contains 768 tasks, organized as 256 tasks for each of 3 curriculum iterations.

## Dataset Columns

Column | Description
--- | ---
`data source` | Environment: `libreoffice_impress`, `libreoffice_calc`, `libreoffice_writer`, `thunderbird`, `Celestia`, or `KAlgebra`.
`iteration` | Curriculum iteration, one of `1`, `2`, or `3`.
`task` | Generated task.

## Load with `datasets`

```python
from datasets import load_dataset

ds = load_dataset(
    "parquet",
    data_files={
        "qwen3vl": "qwen3vl/impress_calc_writer_thunderbird_celestia_kalgebra_iter1_3_tasks.parquet",
        "uitars": "uitars/impress_calc_writer_thunderbird_celestia_kalgebra_iter1_3_tasks.parquet",
    },
)
print(ds["qwen3vl"][0])
```

Read Parquet directly:

```python
import pandas as pd

df = pd.read_parquet("qwen3vl/impress_calc_writer_thunderbird_celestia_kalgebra_iter1_3_tasks.parquet")
print(df.groupby(["data source", "iteration"]).size())
```

## Citation

If this dataset or related resources are useful for your research, please cite ACuRL:

```bibtex
@misc{xue2026autonomouscontinuallearningcomputeruse,
      title={Autonomous Continual Learning of Computer-Use Agents for Environment Adaptation}, 
      author={Tianci Xue and Zeyi Liao and Tianneng Shi and Zilu Wang and Kai Zhang and Dawn Song and Yu Su and Huan Sun},
      year={2026},
      eprint={2602.10356},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.10356}, 
}
```