---
title: AdvGLUE
canonical_url: "https://www.modelscope.cn/datasets/OmniData/AdvGLUE"
md_url: "https://www.modelscope.cn/datasets/OmniData/AdvGLUE.md"
repository: OmniData/AdvGLUE
last_updated: 2024-06-29
license: "[CC BY-SA 4.0]"
storage_size: "45 KB"
domain:
  - publishDate
  - displayName
  - publishUrl
  - paperUrl
  - mediaTypes
  - labelTypes
  - taskTypes
  - publisher
tasks:
  - 2021
  - AdvGLUE
  - "https://adversarialglue.github.io/"
  - "https://arxiv.org/pdf/2111.02840v2.pdf"
  - Text
  - "Visual Question Answering"
  - "Zhejiang University"
downloads: 235
stars: 0
---

# AdvGLUE

> AdvGLUE - OmniData 在 ModelScope 开源的数据集。displayName: AdvGLUE labelTypes: Text license: CC BY-SA 4.0 mediaTypes: Text paperUrl: https://arxiv.org/pdf/2111.02840v2.pdf publishDate: "2021" publishUrl: https://adversarialglue.github.io/ publisher: Microsoft…

OmniData/AdvGLUE 是 ModelScope 魔搭社区上的2021、AdvGLUE、https://adversarialglue.github.io/数据集，涉及 publishDate、displayName、publishUrl 领域，存储大小 45 KB，采用 [CC BY-SA 4.0] 许可。

- **Repository**: OmniData/AdvGLUE
- **License**: [CC BY-SA 4.0]
- **Tasks**: 2021, AdvGLUE, https://adversarialglue.github.io/, https://arxiv.org/pdf/2111.02840v2.pdf, Text, Visual Question Answering, Zhejiang University
- **Domain**: publishDate, displayName, publishUrl, paperUrl, mediaTypes, labelTypes, taskTypes, publisher
- **Storage size**: 45 KB
- **Downloads**: 235
- **Stars**: 0
- **Last updated**: 2024-06-29

Source: https://www.modelscope.cn/datasets/OmniData/AdvGLUE

---

displayName: AdvGLUE
labelTypes:
- Text
license:
- CC BY-SA 4.0
mediaTypes:
- Text
paperUrl: https://arxiv.org/pdf/2111.02840v2.pdf
publishDate: "2021"
publishUrl: https://adversarialglue.github.io/
publisher:
- Microsoft
- University of Illinois Urbana-Champaign
- Zhejiang University
tags:
- Text
taskTypes:
- Open-Domain Question Answering
- Visual Question Answering

---
# 数据集介绍
  ## 简介
  Adversarial GLUE (AdvGLUE) 是一种新的多任务基准，用于定量和深入地探索和评估现代大规模语言模型在各种类型的对抗性攻击下的漏洞。特别是，我们系统地将 14 种文本对抗攻击方法应用于 GLUE 任务以构建 AdvGLUE，并进一步验证了人类的可靠注释。描述来自：对抗性 GLUE：语言模型鲁棒性评估的多任务基准
  ## 引文
  ```
 @article{wang2021adversarial,
 title={Adversarial glue: A multi-task benchmark for robustness evaluation of language models},
 author={Wang, Boxin and Xu, Chejian and Wang, Shuohang and Gan, Zhe and Cheng, Yu and Gao, Jianfeng and Awadallah, Ahmed Hassan and Li, Bo},
 journal={arXiv preprint arXiv:2111.02840},
 year={2021}
 }
 ```
  
## Download dataset
:modelscope-code[]{type="git"}