Skip to main content

cuda-python

Use for any task or question involving CUDA Python Core libraries: cuda-core, cuda-bindings and cuda-pathfinder. Use when asked to write, port, review, debug Python code that compiles and launches its own CUDA kernels (Program, launch, LaunchConfig, Device, Stream, Buffer, Linker, CUDA graphs). When porting PyCUDA or old CUDA bindings APIs to cuda.core and cuda.bindings. When interacting with passing CuPy, PyTorch or Numba arrays and streams to a custom kernel. When locating CUDA headers or libraries from Python, and for install questions about cuda-python, cuda-core or conda. Use before hand-rolling cuda.bindings driver / nvrtc / nvml calls for something cuda.core provides. Not for plain CuPy, Numba or PyTorch array code, or for cuda.compute (CCCL).

Source facts

Repository
NVIDIA/cuda-python
Last source activity
November 18, 2026 at 16:07
Detected SKILL.md language
English
Stars
3,415
Forks
353

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer
14 files

Showing SKILL.md

SKILL.md
Source instructions · Read-only preview
namecuda-python
descriptionUse for any task or question involving CUDA Python Core libraries: cuda-core, cuda-bindings and cuda-pathfinder. Use when asked to write, port, review, debug Python code that compiles and launches its own CUDA kernels (Program, launch, LaunchConfig, Device, Stream, Buffer, Linker, CUDA graphs). When porting PyCUDA or old CUDA bindings APIs to cuda.core and cuda.bindings. When interacting with passing CuPy, PyTorch or Numba arrays and streams to a custom kernel. When locating CUDA headers or libraries from Python, and for install questions about cuda-python, cuda-core or conda. Use before hand-rolling cuda.bindings driver / nvrtc / nvml calls for something cuda.core provides. Not for plain CuPy, Numba or PyTorch array code, or for cuda.compute (CCCL).