dvilasuero/jailbreak-classification-reasoning-models dataset hosted on Hugging Face and contributed by the HF Datasets community

Eval models for classification on your own data

This dataset contains the results of evaluating reasoning models for classification. It contains the pipeline and the code to run it. You can tune the config to run different prompts over your HF datasets.

  Results

Model Accuracy Total Correct Empty

qwq32b-classification 92.00% 100 92 1

r1-classification 91.00% 100 91 2

llama70-classification 77.00% 10077 10

  How to run it

The pipeline uses… See the full description on the dataset page: https://huggingface.co/datasets/dvilasuero/jailbreak-classification-reasoning-eval.

R1-1776 Jailbreaking Examples

The R1-1776 Jailbreaking Examples dataset comprises instances where attempts were made to bypass the safety mechanisms of the R1-1776 model—a version of DeepSeek-R1 fine-tuned by Perplexity AI to eliminate specific censorship while maintaining robust reasoning capabilities. This dataset serves as a resource for analyzing vulnerabilities in language models and developing strategies to enhance their safety and reliability.

  Dataset Summary… See the full description on the dataset page: https://huggingface.co/datasets/weijiejailbreak/r1-1776-jailbreak.

Multimodal Pragmatic Jailbreak on Text-to-image Models

The Multimodal Pragmatic Unsafe Prompts (MPUP) is a dataset designed to assess the multimodal pragmatic safety in Text-to-Image (T2I) models. It comprises two key sections: image_prompt, and text_prompt.

  Dataset Usage







  Downloading the Data

To download the dataset, install Huggingface Datasets and then use the following command: from datasets import load_dataset dataset =… See the full description on the dataset page: https://huggingface.co/datasets/tongliuphysics/multimodalpragmatic.

Results Summary

Model Accuracy Total Correct Unparsable

qwq32 90.00% 100 90 4

r1 93.00% 100 93 3

llama70B 69.00% 100 69 18

  Prediction Distribution

Model Benign Jailbreak Unparsable

qwq32 44 52 4

r1 43 54 3

llama70B 50 32 18

WildJailbreak

Paper: WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models Data: DatasetHF_link

  WildJailbreak Dataset Card

WildJailbreak is an open-source synthetic safety-training dataset with 262K vanilla (direct harmful requests) and adversarial (complex adversarial jailbreaks) prompt-response pairs. In order to mitigate exaggerated safety behaviors, WildJailbreaks provides two contrastive types of queries: 1) harmful… See the full description on the dataset page: https://huggingface.co/datasets/walledai/WildJailbreak.