A public notebook-style collection of CUDA kernels, notes on CUTLASS/CuTe, Triton, PTX, PyTorch internals, and LLM optimization. It aggregates code and notes across multiple directories for GPU systems work.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
A repository that collects hands-on CUDA kernels, CUTLASS/CuTe notes, Triton examples, PTX ISA notes, PyTorch internals notes, and LLM inference/training optimization material. It is described as one of the author’s main public study and engineering notebooks for GPU systems work.
How it works
The repository is organized into multiple subdirectories, including:
cuda-kernels/: handwritten CUDA kernels for reduce, softmax, elementwise, GEMV, indexing, atomic add, upsampling, and linear attention.cuda-mode/: notes and code from the CUDA-MODE lecture series.cutlass/: CUTLASS and CuTe DSL notes, including GEMM, TMA, WGMMA, swizzling, and instruction-level material.triton/: Triton kernels, PyTorch interop examples, and meetup notes.large-language-model/: LLM serving, training, and systems optimization notes.pytorch/: PyTorch internals and CUDA-related notes.papers/: GPU architecture and ML systems paper notes.ptx-isa/: PTX ISA study notes.tools/: small helper scripts.deprecated/: older material kept for reference.
Getting started
The README excerpt does not provide explicit install or run instructions. No command lines are quoted in the provided facts beyond repository structure descriptions.
Recent releases
Latest three releases are image asset migrations labeled as Markdown image assets with dates 2026-08-05. Names:
- mdnice-assets-2026-08-05-3 Markdown image assets (3/3) (2026-08-05)
- mdnice-assets-2026-08-05-2 Markdown image assets (2/3) (2026-08-05)
- mdnice-assets-2026-08-05-1 Markdown image assets (1/3) (2026-08-05)
Traction
Stars: 3186. Forks: 289. Open issues: 6. Language: Cuda. License: none listed. Created: 2018-07-18. Last push: 2026-08-04.
Behind the repo
Linked topics include cuda and llm. No startup or company information is provided in the facts.
Caveats
License is not listed. There are 6 open issues as of the provided snapshot. The project began in 2018, with the last push on 2026-08-04.