This repo contains the code and data for "MEGA-Bench Scaling Multimodal Evaluation to over 500 Real-World Tasks"

A reading list for large models safety, security, and privacy (including Awesome LLM Security, Safety, etc.).

[ICML 2024] TrustLLM: Trustworthiness in Large Language Models

Every practical and proposed defense against prompt injection.

[MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

An easy-to-use LLMs quantization package with user-friendly apis, based on GPTQ algorithm.

A Dynamic Environment to Evaluate Attacks and Defenses for LLM Agents.

This repository provides implementation to formalize and benchmark Prompt Injection attacks and defenses

The official implementation of our pre-print paper "AutoDAN-Turbo: A Lifelong Agent for Strategy Self-Exploration to Jailbreak LLMs".

Official Implementation for EMNLP 2024 (main) "AgentReview: Exploring Academic Peer Review with LLM Agent."

Implementation of paper 'Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference' [NeurIPS'24]

Improving Alignment and Robustness with Circuit Breakers

A trivial programmatic Llama 3 jailbreak. Sorry Zuck!

[ICCV 2023] Consistent Image Synthesis and Editing

HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal

Code for the paper "Evaluating Large Language Models Trained on Code"

Large Language Model guided Protocol Fuzzing (NDSS'24)

O1 Replication Journey: A Strategic Progress Report – Part I

[NeurIPS 2024 Oral🔥] DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs.

Your image is almost there!

[BIBM 2024] XNet v2: Fewer Limitations, Better Results and Greater Universality

A community-maintained Python framework for creating mathematical animations.

Animation engine for explanatory math videos

real time face swap and one-click video deepfake with only a single image

LLM training in simple, raw C/CUDA

A curated list of trustworthy deep learning papers. Daily updating...

Learning from synthetic data - code and models

Cambrian-1 is a family of multimodal LLMs with a vision-centric design.

[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.