Environment Requirements

This document uses openEuler 24.03 LTS SP3 + SGLang 0.5.10 + CUDA 13.0 as an example to describe the installation environment. In addition, this section describes other environments supported by SGLang installed on Kunpeng servers.

Supported Versions

**Table 1** Supported versions
OS	Python	SGLang
openEuler 24.03 LTS SP3	Python 3.11.6	SGLang 0.5.10
openEuler 22.03 LTS SP4	Python 3.9.9	SGLang <= 0.5.0rc0; SGLang 0.5.0rc0 is recommended.

SGLang 0.5.10 requires Python 3.10 or later. The default Python version of openEuler 22.03 LTS SP4 is 3.9.9, which does not meet the version requirements. You are advised to use SGLang 0.5.0rc0 that supports Python 3.9.

Dependent Component Requirements

**Table 2** Version requirements for dependent components
Component	Version
python3	3.11.6-20.oe2403sp3
python3-pip	23.3.1-7.oe2403sp3
pip	26.1.1
PyTorch	2.11.0 CUDA version
CUDA Toolkit	13.0
sglang-kernel	0.4.1
SGLang	0.5.10

To enable the GPU capability, the system must provide an accessible NVIDIA GPU device, CUDA Toolkit 13.0, and CUDA-based PyTorch. The SGLang GPU inference path consists of the main Python package of SGLang and the sglang-kernel CUDA extension. The main Python package of SGLang provides the runtime framework, APIs, and scheduling logic, while the sglang-kernel provides high-performance operator capabilities related to GPU inference. When installing for GPU scenarios, you are advised to install and verify both.

Parent topic: Installation Preparations