The SGLang Ecosystem
A collection of projects built around SGLang for learning, development, and deployment.
LMSYS
The non-profit incubating SGLang alongside Chatbot Arena, FastChat, and Vicuna, co-developing open models, datasets, systems, and evaluation tools.
Miles
An enterprise-facing reinforcement learning framework for large-scale MoE post-training and production workloads.
SGLang-TorchTPU
A PyTorch-native backend for serving LLMs and vision-language models on Google TPUs, built on SGLang and torch_tpu.
SGLang-Jax
A JAX-based inference engine designed for serving LLMs on Google TPUs with high throughput and low latency.
Mini-SGLang
A minimal, educational implementation of SGLang for understanding its core concepts and architecture.
SpecForge
An end-to-end training ecosystem for speculative decoding draft models, natively integrated with SGLang.
Slime
A post-training framework for scaling RL on LLMs, combining high-performance training with flexible data generation.
AReal
An open-source asynchronous RL training system for large reasoning and agentic models.