icon

layout

hat-chef

width

title

description

tableOfContents

outline

pagination

metadata

default

visible
true

visible
true

visible

visible

visible
true

visible

SGLang Cookbook

SGLang offers a wide array of configuration parameters. This manual provides detailed coverage, but what users often need is a straightforward, ready-to-use setup that delivers strong performance, particularly when low latency is essential. The cookbook aim to answer the question:

How do i serve model X in hardware Y?

Step 1 Install SGLang on the hardware Y

	Cover image
NVIDIA H-Series, A-Series, and RTX GPUs	01-nvidia-logo-horiz-500x200-2c50-d.png	nvidia-h-series-a-series-and-rtx-gpus.md
NVIDIA Blackwell/AMD/Intel/NVIDIA Jetson/Ascend hardware	download.jpeg	installation

SGLang's official model releases typically focus on H100 and B200 GPUs. However, the completeness of this cookbook relies on the open source community. We welcome contributions that extend support beyond these hardware specifications. Contributors can follow the format for existing modes, please also attach benchmarks results using our unified benchmark framework genai-bench➚

Step 2 Serve the model X

Model		Cover image
DeepSeek V3.1/V3/R1	deepseek-v3.1-v3-r1	deepseek-logo-icon.png
Meta Llama	meta-llama.md	Llama-2-Model-Details.webp
GPT-OSS	gpt-oss	20250102113948245.png

Step 3 Use the model

{% content-ref url="use-the-model/" %} use-the-model {% endcontent-ref %}