tags:
- embeddings
- text
- security
pretty_name: 'Vigil: LLM Gandalf Instruction Bypass text-embedding-ada-002'
Vigil: LLM Gandalf Instruction Bypass text-embedding-ada-002
Vigil is a Python framework and REST API for assessing Large Language Model (LLM) prompts against a set of scanners to detect prompt injections, jailbreaks, and other potentially risky inputs.
This repository contains text-embedding-ada-002 embeddings for the Lakera Gandalf "Ignore Instructions" dataset.
All prompts from the original dataset have been lowercased before embedding.
You can use the parquet2vdb.py utility to load the embeddings in the Vigil chromadb instance, or use them in your own application.
Format
[
{
"text": str,
"embedding": [],
"model": "text-embedding-ada-002"
}
]
Original dataset: https://huggingface.co/datasets/Lakera/gandalf_ignore_instructions
@InProceedings{gandalf_ignore_instructions,
title = {gandalf_ignore_instructions},
author={Lakera AI (https://www.lakera.ai)},
year={2023}
}