declare-lab/red-instruct

Codes and datasets of the paper Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignment

39
/ 100
Emerging

This project helps evaluate how safely large language models (LLMs) respond to harmful questions, using different prompt styles to test their safety guardrails. You provide a set of potentially harmful questions and specific prompt templates, and it generates model responses. The outcome is a safety score, known as Attack Success Rate (ASR), indicating how easily an LLM can be prompted to give an unsafe answer. This tool is for AI safety researchers and developers who need to rigorously test and improve the safety of their LLMs.

108 stars. No commits in the last 6 months.

Use this if you need to systematically assess and benchmark the safety of various large language models against known harmful queries and red-teaming techniques.

Not ideal if you are looking for a simple, non-technical tool for general content moderation or for testing a single model without deep technical analysis.

AI Safety LLM Evaluation Red Teaming Content Moderation Harmful Content Detection
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 9 / 25
Maturity 16 / 25
Community 14 / 25

How are scores calculated?

Stars

108

Forks

13

Language

Python

License

Apache-2.0

Last pushed

Mar 08, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/declare-lab/red-instruct"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.