Introduction

This document describes how to develop SGLang on Kunpeng servers, including source code build, installation, and basic import verification.

SGLang uses core technologies such as RadixAttention tree cache and compiler-level static graph analysis to reduce repeated prefix computation in multi-turn dialogs and agent tasks by more than 75%, improving throughput by up to 6.4 times. Its open source Arm64 architecture supports seamless migration on the openEuler OS without any code modification.