Attention GitHub Repositories (2026) · Open Source Tools

FlashInfer: Kernel Library for LLM Serving

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

cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.

High Performance KV Cache for LLM Inference

an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch