深度学习框架下接芯片、上承应用，为模型的开发、训练和部署提供系统的支撑，极大地加速人工智能创新研发与应用，其作用类比人工智能时代的操作系统。本文对深度学习框架发展现状、核心技术及生态要素进行深入调研分析，从人工智能产业化需求、软硬件适配与融合优化、大模型快速发展及科学计算前沿交叉领域等不同视角，剖析深度学习框架发展趋势及面临的挑战，并从加速深度学习框架与硬件的适配融合、支持国内主流深度学习框架推广、发挥深度学习框架对AI for Science的支撑作用、加快发展国产深度学习框架生态等方面提出对策建议。
Deep Learning Frameworks serve as a critical bridge between low-level AI hardware and user-facing AI applications by providing essential building blocks for designing，training，and deploying deep learning models. Similar to how operating systems underpin software development in the desktop and mobile era，deep learning frameworks support and drive AI research and innovation. This paper presents a comprehensive survey of the current state of deep learning frameworks，covering their core technologies and ecosystems. It provides an in-depth analysis of the latest trends and challenges in deep learning frameworks，from various perspectives such as industrial applications of AI，hardware-software integration and optimization，recent advances in Large Language Models （LLM），and interdisciplinary frontiers of scientific computing. Furthermore，the paper offers several recommendations to accelerate hardware-software integration for deep learning frameworks，promote the growth of Chinese deep learning frameworks，support cutting-edge initiatives such as AI for Science，and strengthen the Chinese deep learning framework ecosystems.
| Keywords： | Large Language ModelsDeep Learning FrameworksHardware-Software IntegrationScientific ComputingEcosystems Construction |
