Add heavyweight AOT CUDA/C++ kernels to sgl-kernel with tests and benchmarks.
npx skills add https://github.com/sqjian/sglang --skill add-sgl-kernel-sqjian
Please help me install this Agent Skill. Skill: add-sgl-kernel Source: https://github.com/sqjian/sglang/tree/main/.claude/skills/add-sgl-kernel Command: npx skills add https://github.com/sqjian/sglang --skill add-sgl-kernel-sqjian
This guide helps developers add heavyweight AOT CUDA/C++ kernels to the sgl-kernel project, including end-to-end steps for testing and benchmarking.
Follow this guide to implement a new heavyweight AOT CUDA/C++ kernel in sgl-kernel, including tests and benchmarks.
High-intent search queries and answers about installing and using this skill.
To add a new CUDA kernel to a PyTorch extension, implement the kernel in csrc, declare and register the op with the Torch extension, update build scripts, and expose a Python API for testing and benchmarking.
AOT compilation for CUDA kernels involves Ahead-Of-Time compiling C++ and CUDA code into heavyweight extensions before runtime, requiring build script updates and Torch extension registration to integrate with Python APIs.
Benchmark and test a custom CUDA kernel by exposing a Python API that enforces validation, device and dtype dispatch, and performance measurement through integrated unit tests and benchmarks.
Adding a heavyweight CUDA kernel requires updating build scripts to compile the AOT C++ and CUDA code, alongside declaring the op and registering it with the Torch extension for successful integration.
You can use this workflow to accelerate an AI inference pipeline by adding custom matrix-scaling kernels, providing a complete integration flow from C++ implementation to Python API testing and performance benchmarking.
Device and dtype dispatch is needed when integrating a CUDA kernel to enforce complete validation patterns, ensuring the PyTorch extension correctly handles diverse tensor types and hardware targets during execution.
hey-kongImplement and integrate CUDA/C++ kernels with tests and benchmarks.
spacecat2002Add AOT CUDA or C++ kernels to sgl-kernel with CMake and pytest.
guqiong96Add custom CUDA/C++ kernels to the sgl-kernel library with CMake and Python API integration.
annealing-inversionAdd AOT CUDA/C++ kernels to sgl-kernel with tests and benchmarks.
