--- title: Welcome to SGLang description: High-performance serving framework for large language and multimodal models. keywords: - sglang - llm serving - multimodal - inference runtime mode: wide --- Star Fork

Designed for low-latency, high-throughput inference with RadixAttention, prefix caching, and multi-GPU parallelism. Broad support for Llama, Qwen, DeepSeek, and more. Compatible with Hugging Face and OpenAI APIs. Native support across Hardware Platforms including NVIDIA, AMD, Intel Xeon, Google TPU, and Ascend NPU accelerators. Open-source with widespread adoption, powering 400k+ GPUs and integrated with major RL frameworks. SGLang powers large-scale production deployments, generating trillions of tokens each day across more than 400,000 GPUs worldwide. It is hosted under the non-profit open-source organization [LMSYS](https://lmsys.org/about/). --- ## Get Started SGLang is an inference framework meant for production level serving. It is designed to deliver low-latency and high-throughput inference across a wide range of setups, from a single GPU to large distributed clusters. Install SGLang with pip, from source, or via Docker on your preferred hardware platform. Launch your first model server and send requests in minutes with OpenAI-compatible APIs. ## News and latest blogs {/* BEGIN_LMSYS_SGLANG_BLOG_CARDS */}
DSpark in SGLang: Speculative Decoding with Confidence-Driven, Variable-Length Verification

{"DSpark in SGLang: Speculative Decoding with Confidence-Driven, Variable-Length Verification"}

{"July 6, 2026"}

Agent-Assisted SGLang Development: An Initial Exploration

{"Agent-Assisted SGLang Development: An Initial Exploration"}

{"July 2, 2026"}

Improving DeepEP MoE Load Balance in SGLang with Waterfill and LPLB

{"Improving DeepEP MoE Load Balance in SGLang with Waterfill and LPLB"}

{"June 26, 2026"}

MOSS-TTS Local Transformer v1.5 on SGLang-Omni: Serving Native-Streaming 48 kHz Speech

{"MOSS-TTS Local Transformer v1.5 on SGLang-Omni: Serving Native-Streaming 48 kHz Speech"}

{"June 17, 2026"}

Optimizing Ling-2.6-1T on TPU with SGLang-JAX: Hiding MoE Data Movement Behind Compute with One Pallas Kernel

{"Optimizing Ling-2.6-1T on TPU with SGLang-JAX: Hiding MoE Data Movement Behind Compute with One Pallas Kernel"}

{"June 17, 2026"}

The next generation of speculative decoding: DFlash and Spec V2

{"The next generation of speculative decoding: DFlash and Spec V2"}

{"June 15, 2026"}

{/* END_LMSYS_SGLANG_BLOG_CARDS */} --- ## Learn more and join the community

Stay connected

{" "} Development roadmap to follow current priorities and upcoming work.
{" "} Weekly public development meeting to hear updates and join open discussions.
{" "} Slack for questions, feedback, and community support.
X Twitter and {" "} LinkedIn for project updates.
{" "} LMSYS blog for release notes, benchmarks, and technical deep dives.
{" "} Learning materials for blogs, slides, and videos.