SGLang Project
To accelerate AI model inference and reduce compute costs through high-performance optimization and open-source systems.
At a Glance
- AI Labs
- Cloud Service Providers
- Enterprise AI Developers
- Open-source Community
AI Tools by SGLang Project
(1)
SGLang
LLM and VLM Serving Framework
Discussions
No discussions yet
Be the first to start a discussion about SGLang Project
Latest News
Sources: Project SGLang spins out as RadixArk with $400M valuation
Introducing Miles — RL Framework To Fire Up Large-Scale MoE Training
ROCm Support for Miles: Large-Scale RL Post-Training on AMD
SGLang reaches 400,000 GPUs deployment milestone
Products & Services
A high-performance open-source serving framework for large language models (LLMs) and vision language models (VLMs).
An enterprise-facing reinforcement learning framework tailored for large-scale Mixture-of-Experts (MoE) training and production.
A paid hosting service for enterprises to deploy SGLang with production-grade reliability and low latency.
Market Position
RadixArk positions itself as a higher-performance alternative to vLLM and HuggingFace TGI, focusing on memory efficiency (RadixAttention) and enterprise RL workflows (Miles).
Leadership
Founders
Ying Sheng
Co-founder and CEO of RadixArk. Previously an engineer at Elon Musk's AI startup xAI and a research scientist at Databricks. Key contributor to the SGLang project. Stanford University graduate/Ph.D. candidate.
Lianmin Zheng
Co-founder. Ph.D. from UC Berkeley advised by Ion Stoica. Lead of LMSYS Org projects including FastChat and Vicuna. Co-creator of SGLang.
Ion Stoica
Co-founder and Chairman/Advisor. Professor of Computer Science at UC Berkeley. Co-founder of Databricks, Anyscale, and Conviva.
Executive Team
Ying Sheng
Co-founder and CEO
Former engineer at xAI and research scientist at Databricks; SGLang maintainer.
Lianmin Zheng
Co-founder
Ph.D. from UC Berkeley; lead of LMSYS Org and creator of Vicuna.
Board of Directors
Founding Story
RadixArk originated as SGLang in 2023 within the UC Berkeley lab of Databricks co-founder Ion Stoica. The project was created to address the inefficiencies in LLM inference. Following massive community adoption and deployment on over 400,000 GPUs, the core contributors spun the project out into a commercial entity to provide enterprise-grade infrastructure and managed services.
Business Model
Revenue Model
Open-core model: Free open-source inference engine (SGLang) and RL framework (Miles) with revenue generated through managed hosting services and enterprise support.
Pricing Tiers
Access to core SGLang and Miles code via GitHub.
Production-grade hosting for SGLang inference.
Target Markets
- AI Labs
- Cloud Service Providers
- Enterprise AI Developers
- Open-source Community
- High-throughput LLM serving
- Real-time chat applications
- Large-scale RL post-training
- Multi-modal model inference
- xAI
- Cursor
- LMSYS Org