Add a lightweight JIT CUDA kernel to the sglang framework.
npx skills add https://github.com/rayleizhu/sglang --skill add-jit-kernel
Please help me install this Agent Skill. Skill: add-jit-kernel Source: https://github.com/rayleizhu/sglang/tree/main/.claude/skills/add-jit-kernel Command: npx skills add https://github.com/rayleizhu/sglang --skill add-jit-kernel
This Skill provides a structured, step-by-step guide to integrate new, lightweight CUDA kernels into the SGLang framework, streamlining the development of custom high-performance operations.
Follow the tutorial to implement a new JIT kernel by creating the necessary C++ and CUDA files and adding a Python wrapper.
High-intent search queries and answers about installing and using this skill.
To add custom CUDA kernels to SGLang, you implement the kernel using Just-In-Time (JIT) compilation, create a Python wrapper, tune build flags, and run tests and benchmarks.
JIT compilation in SGLang allows you to register and execute custom high-performance tensor operations by dynamically compiling CUDA device code during runtime.
Yes, SGLang provides common C++ and CUDA abstractions for host-side utilities, device-side utilities, tensor validation, type traits, and vectorized memory access to ensure safer kernel development.
The best way to integrate custom activation functions is to follow the structured guide for adding lightweight JIT CUDA kernels, which covers implementation, Python wrapper creation, and build flag tuning.
Yes, the process of adding a JIT kernel involves tuning build flags to ensure your custom CUDA kernel compiles and executes correctly within the SGLang framework.
Using JIT compilation for custom element-wise transformations streamlines the integration of specialized high-performance operations directly into SGLang, bypassing the need for static framework recompilation.
sgl-projectGuide adding lightweight JIT CUDA kernels to the sglang framework.
coreyjaDraft SKILL.md frontmatter and structure guidance for Claude Code skills.
sqjianAutomate adding lightweight JIT CUDA kernels to SGLang's jit_kernel module.
guqiong96Develop and optimize Triton and CUDA JIT kernels for diffusion models.
