HackAPrompt Dataset
Prompt based jailbreak attempts, for testing model vulnerability.
About HackAPrompt Dataset
The dataset is published on Hugging Face under the identifier hackaprompt/hackaprompt-dataset. This dataset contains submissions from a prompt hacking competition. Submissions were sourced from two environments: a playground for experimentation and an official submissions platform.
The dataset is meant to be used in a research context to better understand the different types of attacks "in the wild" on LLMs. An in-depth analysis of the dataset has been accepted at the EMNLP 2023 conference.
Drafted from huggingface.co · not yet confirmed by Learn Prompting
How it works
HackAPrompt is the dataset from the largest prompt hacking competition, with over 600,000 attempts to make models ignore instructions, leak system prompts, or produce a target string. Each record has the prompt, the model, the challenge level and whether it succeeded.
It's the largest open corpus of prompt injection attempts and the standard training data for injection detectors.
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing HackAPrompt Dataset? Send them this page.