OSS Alternative - Discover Top Open Source Alternatives to Popular Software

PKU-Alignment/align-anything

A modular framework for aligning any-modality large models with human intentions and values using diverse fine-tuning and reinforcement learning methods.

Core Features

Highly modular framework for easy customization and modification.

Supports fine-tuning of diverse multi-modal models (image/video/audio).

Implements various alignment algorithms (SFT, DPO, PPO, Rule-based RL).

Provides a multi-modal command-line interface (CLI).

Accelerated training with vLLM engine and NPU support.

Quick Start

pip install align-anything

Detailed Introduction

Align-Anything is an advanced open-source framework developed by PKU-Alignment Team, designed to facilitate the alignment of any-modality large models (any-to-any models) with human intentions and values. It offers a highly modular architecture, enabling researchers and developers to easily customize and apply various fine-tuning and reinforcement learning algorithms, including SFT, DPO, and PPO, across diverse multi-modal data like images, videos, and audio. The project emphasizes efficiency with features like vLLM integration and support for both Nvidia GPUs and Huawei Ascend NPUs, making it a comprehensive toolkit for cutting-edge AI alignment research and development.