Guide adding AOT CUDA/C++ kernels to the sgl-kernel library.
npx skills add https://github.com/sgl-project/sglang --skill add-sgl-kernel-sgl-project
Please help me install this Agent Skill. Skill: add-sgl-kernel Source: https://github.com/sgl-project/sglang/tree/main/.claude/skills/add-sgl-kernel Command: npx skills add https://github.com/sgl-project/sglang --skill add-sgl-kernel-sgl-project
This Skill provides a comprehensive guide to integrating custom, heavyweight CUDA/C++ kernels into the sgl-kernel library, ensuring they are tested and benchmarked.
Follow the tutorial to implement, test, and build a new CUDA kernel for sgl-kernel.
High-intent search queries and answers about installing and using this skill.
To add a custom CUDA kernel to a PyTorch C++ extension, you write the kernel in C++, modify C++ declarations, register the Torch extension, update CMake build files, and expose the Python API to complete the integration.
Integrating AOT C++ kernels into sgl-kernel requires implementing the kernel code, updating C++ declarations, registering the Torch extension, configuring the CMake build system, exposing the Python API, and writing tests and benchmarks.
You benchmark custom CUDA kernels using pytest and Triton by creating dedicated performance benchmark suites alongside unit tests. This ensures the newly integrated operations meet performance requirements within the library.
Yes, modifying the CMake build system is required to expose new Python APIs for sgl-kernel. The integration workflow dictates updating CMake build files to correctly compile and link the custom C++ operations into the PyTorch extension.
Yes, you can use CUTLASS for performance-critical tensor operations in sgl-kernel. The Skill specifically guides developers through implementing heavy-weight operations that rely on CUTLASS from kernel code to final Python API exposure.
spacecat2002Add AOT CUDA or C++ kernels to sgl-kernel with CMake and pytest.
annealing-inversionAdd AOT CUDA/C++ kernels to sgl-kernel with tests and benchmarks.
scottgl9Implement and integrate CUDA/C++ kernels into sgl-kernel for tensor operations.
moirai-internalAdds custom CUDA/C++ kernels to the sgl-kernel library with PyTorch extension integration and CMake build system support.
