#5252nd of 126 in ML frameworks and libraries
cutlass
CUDA C++ templates and Python DSL for writing GEMM and tensor kernels
- Stars
- 10.7k
- License
- custom license
- Last commit
- Oct 2026
- Last release
- Oct 2026
- Language
- C++
9,472 on 10 Apr 202610,684 on 30 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
CUTLASS is a library of CUDA C++ template abstractions for building matrix-multiply (GEMM) and related kernels, with CuTe layout and tensor primitives at its core. Version 4 adds CuTe DSL, a Python interface (public beta) for writing kernels against the same concepts with faster compile times. It covers FP64 down to FP4 and block-scaled types on Volta through Blackwell, with initial Rubin (SM107) support.
Who it is for: GPU kernel engineers writing custom CUDA GEMM and tensor-core kernels
Strengths
- Supports FP64, FP32, TF32, FP16, BF16, FP8, NVFP4, MXFP4/6/8 and integer types
- CuTe DSL lets kernels be written in Python without deep C++ template expertise
- Covers Volta, Turing, Ampere, Ada, Hopper and Blackwell, with initial Rubin support
- Tested against FlashAttention, FlashInfer, PyTorch and TensorRT-LLM
Weaknesses
- NVIDIA GPUs only; requires CUDA Toolkit 11.4 or newer and a C++17 compiler
- CuTe DSL is in public beta, and the extension compiler is an opt-in preview
- Rubin kernels need the R615 driver; some packed ops miscompile on Rubin
- Low-level kernel library, not a ready-to-use inference or training tool
What it needs
- GPU required
- Needs CUDA Toolkit 11.4+ (12.8 recommended), C++17 compiler, NVIDIA GPU, compute capability 7.0+
Quick answers
Is cutlass open source?
Its code is public on GitHub under its own license terms; read them before you use it at work.
Does cutlass need a GPU?
Yes, it needs a GPU.
Also in ML frameworks and libraries
See all 126| Rank | Project | Score |
|---|---|---|
| 1 | pytorchPython tensor library with GPU support and tape-based autograd for deep learning | 95 out of 100 |
| 2 | transformersPython library defining open models for text, vision, audio and multimodal tasks | 94 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 93 out of 100 |
| #4 | pytorch-image-modelsPyTorch library of image models, pretrained weights and reference training scripts | 93 out of 100 |
| #5 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 91 out of 100 |