modelscope/ms-swift
A scalable and lightweight infrastructure for fine-tuning, inference, and deployment of over 1000 large language models (LLMs) and multimodal large language models (MLLMs) using advanced techniques.
Core Features
Quick Start
pip install ms-swiftDetailed Introduction
ms-swift is a comprehensive framework from the ModelScope community designed for the entire lifecycle of large language models and multimodal large models. It enables efficient training, inference, evaluation, quantization, and deployment for a vast array of models, supporting over 1000 different architectures. The framework incorporates cutting-edge training technologies, including various PEFT methods, reinforcement learning algorithms like GRPO, and advanced distributed training techniques. It also features robust memory optimization and multimodal data handling, making it a versatile tool for developing and deploying intelligent AI applications across diverse hardware environments.