| 引用本文: | 林泓宇,刘洋,罗浩然,李林海,曹行行,王梦娜,徐家豪,李泉毅,张洪滨,邢明杰,武延军.深度学习编译优化技术综述.软件学报,,():1-31 |
| |
|
| |
|
|
| 本文已被：浏览 77次 下载 45次 |
码上扫一扫！ |
|
|
| 深度学习编译优化技术综述 |
|
林泓宇1,2, 刘洋1,2, 罗浩然3, 李林海1,2,4,5, 曹行行1,2,4,5, 王梦娜1,2, 徐家豪1,2, 李泉毅1,2,4,5, 张洪滨1,2, 邢明杰1,2, 武延军1,2
|
|
1.中国科学院 软件研究所 智能软件研究中心, 北京 100190;2.中国科学院大学 计算机科学与技术学院, 北京 100049;3.College of Computing and Data Science, Nanyang Technological University, Singapore 639798, Singapore;4.中国科学院大学南京学院 信息学院, 江苏 南京 211135;5.中科南京软件技术研究院 科学技术部, 江苏 南京 211135
|
|
| 摘要: |
| 随着深度学习的快速发展, 深度学习框架和硬件相关的研究成为推动其发展的重要方向, 其中框架为开发者提供了便捷的深度学习模型构建方式, 而硬件则提供了强大的计算能力. 然而, 当前多样化的深度学习框架与硬件平台之间的适配性差、兼容性不足, 导致了性能和可扩展性问题. 为解决这一挑战, 深度学习编译技术应运而生, 它通过一系列的中间表示及其对应的自动化转换, 将不同框架的模型高效地映射到特定后端设备上的可执行代码, 并在此过程中应用多种深度学习编译优化技术, 如算子融合、内存优化、自动调优等, 在解决可拓展性问题的同时, 显著提升模型的计算效率. 首先梳理深度学习编译的基本概念和整体流程, 随后总结并分类深度学习编译中的常见优化技术, 最后探讨该领域面临的挑战以及未来发展方向. |
| 关键词: 深度学习 深度学习编译 编译优化 大语言模型编译 高性能计算 RISC-V |
| DOI：10.13328/j.cnki.jos.007649 |
| 分类号:TP314 |
| 基金项目: |
|
| Survey on Deep Learning Compilation Optimization Technologies |
|
LIN Hong-Yu1,2, LIU Yang1,2, LUO Hao-Ran3, LI Lin-Hai1,2,4,5, CAO Hang-Hang1,2,4,5, WANG Meng-Na1,2, XU Jia-Hao1,2, LI Quan-Yi1,2,4,5, ZHANG Hong-Bin1,2, XING Ming-Jie1,2, WU Yan-Jun1,2
|
|
1.Intelligent Software Research Center, Institute of Software, Chinese Academy of Sciences, Beijing 100190, China;2.School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing 100049, China;3.College of Computing and Data Science, Nanyang Technological University, Singapore 639798, Singapore;4.School of Information Science, University of Chinese Academy of Sciences, Nanjing, Nanjing 211135, China;5.Department of Science and Technology, Nanjing Institute of Software Technology, Nanjing 211135, China
|
| Abstract: |
| With the rapid development of deep learning, research on deep learning frameworks and hardware has become a crucial direction for advancing the field. Frameworks provide developers with convenient tools for building deep learning models, while hardware delivers powerful computational capabilities. However, limited adaptability and insufficient compatibility between diverse deep learning frameworks and hardware platforms often lead to performance and scalability issues. To address this challenge, deep learning compilation technology has emerged. Models from different frameworks are efficiently mapped to executable code for specific backend devices through a series of intermediate representations and automated transformations. During this process, various deep learning compilation optimization techniques, such as operator fusion, memory optimization, and auto-tuning, are applied. These not only resolve scalability issues but also significantly improve the computational efficiency of the models. This study first reviews the basic concepts and overall process of deep learning compilation, then summarizes and categorizes the common optimization techniques in deep learning compilation, and finally discusses the challenges faced by the field and potential future development directions. |
| Key words: deep learning (DL) deep learning compilation compiler optimization large language model (LLM) compilation high performance computing RISC-V |
|
|
|
|
