# SGLang

SGLang is a high-performance serving framework for large language models and multimodal models.

Repository: https://github.com/sgl-project/sglang
Canonical: https://ross.abutalabs.com/products/sglang
Homepage: https://sglang.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: cuda, inference, llama, llm, moe, transformer, vlm, deepseek, blackwell, gpt-oss, diffusion, attention, glm, minimax, qwen, qwen-image, reinforcement-learning, wan
Last push: 2026-08-27T00:34:46+00:00
Link (homepage): https://sglang.io
Link (site_page): https://docs.sglang.io/

## Health v2 (maintenance only)
Score: 93/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 69
- inputs: {"age_days": 968, "days_push": 7, "days_rel": 12, "gap_med": 13, "n_releases_24m": 46}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 32504, forks 8238 (observed 2026-08-28T04:11:57.580060+00:00)

## What it is
SGLang is a high-performance open-source serving framework for large language models and multimodal models, offering low-latency, high-throughput inference from a single GPU to distributed clusters. It provides OpenAI-compatible APIs, RadixAttention prefix caching, speculative decoding, and broad hardware support including NVIDIA, AMD, TPU, and Ascend accelerators.

## Use cases
- serve an llm with an openai-compatible api
- deploy a self-hosted inference server for llama or qwen
- run high-throughput batch inference on gpus
- serve vision-language models
- speed up chatbot serving with prefix caching
- run rl training rollouts with fast inference
- serve deepseek or glm models on multi-gpu clusters

## When to choose
- you need production-grade, low-latency LLM serving with high throughput
- you want OpenAI-compatible endpoints for self-hosted open models
- you need multi-GPU or multi-node tensor parallelism and disaggregated prefill/decode
- you serve agentic or RAG workloads with heavy shared prefixes
- you need day-0 support for the latest open models

## When to avoid
- you only need to run inference locally on a laptop without GPUs
- you want a simple single-file inference script rather than a server
- your models are unsupported architectures outside its supported list
- you need a managed cloud service rather than self-hosted infrastructure

## Facets
- artifact type: framework
- maturity: stable
- function: llm-inference, http-server, api-framework, machine-learning, gpu-computing, rag
- domain: large-language-models, machine-learning, deep-learning, gpu-computing, apis, self-hosted
- platform: python, cloud, self-hosted
- tags: serving, inference-engine, openai-compatible-api, radixattention, prefix-caching, speculative-decoding, multimodal, vllm-alternative, tensor-parallelism, diffusion-models, linux, docker, gpu

## Member repositories
- sgl-project/sglang (main) score 93
- sgl-project/mini-sglang (mirror) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:11:57.580060+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T16:51:56.569742+00:00, confidence not recorded.
  - readme: https://github.com/sgl-project/sglang (fetched 2026-08-28T04:11:57.580060+00:00, sha 0fca2503d20d)
  - homepage: https://sglang.io (fetched 2026-08-29T07:49:02.011405+00:00, sha 8ecaada3c97f)
  - site_page: https://docs.sglang.io/ (fetched 2026-08-29T07:49:02.021563+00:00, sha d58f15a9063a)
  - registry_pypi: https://pypi.org/pypi/sglang/json (fetched 2026-08-29T07:49:02.024128+00:00, sha a2b61cefeba7)
- Data as of 2026-08-30T08:39:29.467469+00:00.