Dataset assetOpen Source CommunityText ClassificationRisk Assessment

walledai/HarmBench

The dataset comprises three configurations: contextual, copyright, and standard. Each configuration has specific features and splits. The contextual configuration includes `prompt`, `context`, and `category` fields; the copyright configuration includes `prompt` and `tags`; the standard configuration includes `prompt` and `category`. Training set sizes and sample counts differ for each configuration.

Source

hugging_face

Created

Nov 28, 2025

Updated

Jul 31, 2024

Signals

195 views

Availability

Linked source ready

Overview

Dataset description and usage context

Dataset Overview

Dataset Configurations

Configuration Name: contextual

Features:
- prompt: string
- context: string
- category: string
Splits:
- train:
  - Bytes: 45538.0
  - Samples: 100
Download Size: 90186
Dataset Size: 45538.0
Data File Paths:
- train: contextual/train-*

Configuration Name: copyright

Features:
- prompt: string
- tags: string
Splits:
- train:
  - Bytes: 10260.0
  - Samples: 100
Download Size: 4952
Dataset Size: 10260.0
Data File Paths:
- train: copyright/train-*

Configuration Name: standard

Features:
- prompt: string
- category: string
Splits:
- train:
  - Bytes: 22431.5
  - Samples: 200
Download Size: 12347
Dataset Size: 22431.5
Data File Paths:
- train: standard/train-*

Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio