Write GPU kernels with NVIDIA CuTe DSL using Python decorators.
npx skills add https://github.com/NVIDIA/skills --skill kernel-cute-writing
Please help me install this Agent Skill. Skill: kernel-cute-writing Source: https://github.com/NVIDIA/skills/tree/main/skills/TensorRT-LLM/kernel-cute-writing Command: npx skills add https://github.com/NVIDIA/skills --skill kernel-cute-writing
Write and implement GPU kernels using NVIDIA CuTe DSL (CUTLASS 4.x Python API) — NOT for Triton, CUDA C++, or conceptual explanations. Trigger only when the user wants to write or implement a kernel, not when asking questions about CuTe DSL concepts or layouts. CuTe DSL uses cute.jit/cute.kernel decorators and cutlass.cute imports. Covers element-wise kernels, GEMM patterns, reductions, memory hierarchy.
Install the CuTe DSL, implement a kernel with @cute.kernel, and run a host wrapper to verify.
High-intent search queries and answers about installing and using this skill.
Author GPU kernels in CuTe DSL by decorating Python functions with @cute.kernel and @cute.jit decorators. This workflow supports element-wise operations, GEMM-style computations, and reductions within the CUTLASS 4.x Python API framework.
Integrate CuTe DSL kernels with PyTorch by using from_dlpack-based tensor conversion to pass tensors between frameworks. This enables building host wrappers for testing and deploying custom GPU kernels within PyTorch workflows.
Reduce JIT compilation latency for CuTe DSL kernels by pre-compiling them with cute.compile. Combined with dynamic layout marking, this approach minimizes runtime overhead and ensures low JIT latency during GPU kernel execution.
CuTe DSL supports element-wise kernels, GEMM-style computations, and reductions for GPU kernel development. The framework provides patterns for managing memory hierarchy and layouts, covering the core computational patterns needed for high-performance GPU programming.
CuTe DSL works with JAX for GPU kernel testing and deployment through DLPack-based tensor conversion. This allows you to build host wrappers that connect custom kernels with JAX workflows for verification and execution.
You do not need CUDA C++ knowledge to use CuTe DSL for kernel writing. The framework allows you to author high-performance GPU kernels entirely in Python using decorators, bypassing the need for direct CUDA C++ programming.
LancerLabTune and validate Python JIT GPU kernels using CuTe DSL.
mindspore-aiCompile and call CUDA C kernels from PyTorch using load_inline.
yo-stevenWrite and implement GPU kernels using NVIDIA CuTe DSL for element-wise, GEMM, and reduction operations.
fmh66Benchmark custom GPU kernels against PyTorch baselines with CUDA event timing.
