Add a JIT CUDA kernel to SGLang's jit_kernel module with Python wrapper and tests.
npx skills add https://github.com/hey-kong/sglang --skill add-jit-kernel-hey-kong
Please help me install this Agent Skill. Skill: add-jit-kernel Source: https://github.com/hey-kong/sglang/tree/main/.claude/skills/add-jit-kernel Command: npx skills add https://github.com/hey-kong/sglang --skill add-jit-kernel-hey-kong
This guide provides a practical, end-to-end workflow to add a lightweight JIT CUDA kernel to SGLang's jit_kernel module, accelerating development and testing of GPU kernels.
Create a new CUDA kernel, wrap it in Python, and run the included tests and benchmarks.
High-intent search queries and answers about installing and using this skill.
To add a custom CUDA kernel to SGLang, you create the CUDA code under the jit_kernel/csrc directory, expose it through a Python wrapper, and integrate it with the testing and benchmark framework to measure throughput.
The best way to validate inputs and safely launch a JIT CUDA kernel is to use TensorMatcher for input validation and LaunchKernel for robust kernel launches within your Python wrapper integration.
Yes, SGLang supports JIT compilation for performance-focused GPU kernels, enabling lightweight development and testing of elementwise computations and tensor operations directly within the jit_kernel module.
You benchmark a newly integrated CUDA kernel in SGLang by utilizing the included unit and benchmark test scaffolding, which helps measure throughput and evaluate vectorization decisions and optional PDL usage.
Yes, you can use SGLang's jit_kernel module for elementwise computations, applying vectorization decisions and optional PDL usage to optimize performance for GPU-based tensor operations.
FutureMLS-LabAdd validated JIT CUDA kernels to SGLang's jit_kernel module.
OpenMOSSAdd a lightweight JIT CUDA kernel with host wrapper, tests, and benchmarks to SGLang's jit_kernel module.
sqjianAutomate adding lightweight JIT CUDA kernels to SGLang's jit_kernel module.
guqiong96Develop and optimize Triton and CUDA JIT kernels for diffusion models.
