nvidia-cutlass-dsl-libs-base
NVIDIA CUTLASS Python DSL
Decision gist · record as of 2026-08-14
Yes, if you are targeting NVIDIA Tensor Cores on Ampere/Hopper/Blackwell GPUs and want to prototype or deploy optimized kernels from Python. The active maintenance, zero known vulnerabilities, and large download volume indicate real adoption. However, verify the unclear license terms before production use, and be aware the package is in public beta—expect potential API changes before summer 2026 graduation.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires NVIDIA CUDA 13 environment, Linux (manylinux_2_28), and Python 3.10+.
- Wheels available only for x86_64 and aarch64 architectures.
- Medium install friction due to platform-specific wheels (manylinux_2_28, aarch64/x86_64 only) and multiple compiled dependencies including cuda-python and nvidia-cuda-nvdisasm.
License · maintenance · safety
(unclear) — License treatment is unclear—no SPDX identifier or raw license text provided. Verify licensing terms with NVIDIA before using in production or proprietary projects.
last release 2026-08-05 (9 days) · last repo commit 2026-08-14 · 10,250 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 6,269,249 downloads/mo, #1,944 on PyPI
Alternatives
Verify before relying
pip install nvidia-cutlass-dsl-libs-base
import cutlass
# Access CuTe DSL for kernel programming via CUTLASS Python interfaces- Exact API surface and available DSL constructs beyond CuTe DSL mentioned in description
- Performance benchmarks or comparison to C++ CUTLASS implementations
- Timeline and stability guarantees for beta-to-production graduation (stated as summer 2026)
- Compatibility with specific DL frameworks (PyTorch, TensorFlow, etc.) mentioned in description
What it is and what it does
CUTLASS DSL provides a Python-native layer for writing optimized CUDA kernels without deep C++ expertise. The package exposes core CUTLASS and CuTe concepts—layouts, tensors, hardware atoms, and thread/data hierarchy control—allowing developers to target NVIDIA's Tensor Cores on modern GPU architectures. It aims to reduce the learning curve for GPU programming, speed up kernel prototyping, and integrate directly with deep learning frameworks.
The package depends on numpy, protobuf, cuda-python, and nvidia-cuda-nvdisasm, plus a core library (nvidia-cutlass-dsl-libs-core). It is currently in public beta and runs on Linux with Python 3.10 or later. Installation requires platform-specific wheels (x86_64 or aarch64), and a CUDA 13 environment is needed at runtime.
Use it for
- Prototype and optimize matrix multiply kernels targeting Tensor Cores without writing C++
- Develop custom linear algebra operations for deep learning workloads with native framework integration
- Learn GPU programming and CUTLASS concepts with lower barrier to entry than C++ implementations
- Rapidly iterate on kernel designs for performance engineering and research on modern NVIDIA GPUs
- Integrate optimized CUDA kernels into production pipelines without glue code between Python and C++
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, if you are targeting NVIDIA Tensor Cores on Ampere/Hopper/Blackwell GPUs and want to prototype or deploy optimized kernels from Python.
The active maintenance, zero known vulnerabilities, and large download volume indicate real adoption. However, verify the unclear license terms before production use, and be aware the package is in public beta—expect potential API changes before summer 2026 graduation.
Install
nvidia-cutlass-dsl-libs-base on PyPI
Before you install
Medium install friction due to platform-specific wheels (manylinux_2_28, aarch64/x86_64 only) and multiple compiled dependencies including cuda-python and nvidia-cuda-nvdisasm. Package is actively maintained with recent releases, but remains in public beta.
Requires NVIDIA CUDA 13 environment, Linux (manylinux_2_28), and Python 3.10+. Wheels available only for x86_64 and aarch64 architectures.
License in practice
License treatment is unclear—no SPDX identifier or raw license text provided. Verify licensing terms with NVIDIA before using in production or proprietary projects.
Quickstart
pip install nvidia-cutlass-dsl-libs-base
import cutlass
# Access CuTe DSL for kernel programming via CUTLASS Python interfacesVerify before relying
- Exact API surface and available DSL constructs beyond CuTe DSL mentioned in description
- Performance benchmarks or comparison to C++ CUTLASS implementations
- Timeline and stability guarantees for beta-to-production graduation (stated as summer 2026)
- Compatibility with specific DL frameworks (PyTorch, TensorFlow, etc.) mentioned in description
Package facts
| License | Not declared unclear |
| Python support | Supports the current Python release >=3.10 |
| Install friction | Medium. Platform-specific wheel |
| Runtime dependencies | 7 packagesnumpytyping-extensionscuda-pythonbackports.strenumprotobufnvidia-cuda-nvdisasmnvidia-cutlass-dsl-libs-core |
| Maintenance | Actively maintained 9 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 6,269,249 / month, #1,944 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
| Classifiers | Development Status :: 4 - BetaEnvironment :: GPU :: NVIDIA CUDA :: 13License :: Other/Proprietary LicenseOperating System :: POSIX :: LinuxProgramming Language :: Python :: 3 :: OnlyProgramming Language :: Python :: 3.10Programming Language :: Python :: 3.11Programming Language :: Python :: 3.12Programming Language :: Python :: 3.13Programming Language :: Python :: 3.14Programming Language :: Python :: Implementation :: CPython |
Evidence: nvidia_cutlass_dsl_libs_base-4.7.0-cp310-cp310-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp310-cp310-manylinux_2_28_x86_64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp311-cp311-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp311-cp311-manylinux_2_28_x86_64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp312-cp312-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp312-cp312-manylinux_2_28_x86_64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp313-cp313-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp313-cp313-manylinux_2_28_x86_64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp314-cp314-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp314-cp314-manylinux_2_28_x86_64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp314-cp314t-manylinux_2_28_aarch64.whl; nvidia_cutlass_dsl_libs_base-4.7.0-cp314-cp314t-manylinux_2_28_x86_64.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “python cuda kernel programming”
- nvidia-cutlass-dsl-libs-baseProvides Python interfaces for writing high-performance CUDA kernels…
- nvidia-cutlass-dsl-libs-cu12Provides a Python DSL for writing high-performance CUDA kernels using…
- cuda-tilecuTile Python is a compiler and programming language for writing GPU…
Give your agent the search over MCP, or paste the wish link into any chat.
More Scientific/Engineering packages
NumPy provides an N-dimensional array object and a comprehensive suite of mathematical, linear algebra, Fourier transform, and random number functions for scientific computing in Python.
pandas provides fast, flexible data structures (Series and DataFrame) for loading, cleaning, transforming, and analyzing labeled or relational data in Python.
scipy provides numerical algorithms for mathematics, science, and engineering—including optimization, integration, linear algebra, Fourier transforms, signal and image processing, and ODE solvers—built on numpy arrays.
scikit-learn provides a comprehensive Python library for supervised and unsupervised machine learning, including classification, regression, clustering, dimensionality reduction, and model evaluation tools built on NumPy and SciPy.
Install it if you need to train, evaluate, or deploy supervised or unsupervised learning models.
dill extends Python's pickle module to serialize and deserialize a much wider range of Python objects, including functions, lambdas, classes, and interpreter sessions, to byte streams for storage or network transmission.
Multiprocess is an enhanced fork of Python's standard multiprocessing library that uses dill for better serialization, allowing you to spawn processes with a threading-like API and share complex objects between them.
Install it if you use multiprocessing and encounter pickle serialization limits with lambdas or complex objects.
See also apache-tvm-ffi · flydsl · nvidia-cutlass-dsl-libs-core · nvidia-cutlass-dsl-libs-cu12 · nvidia-cutlass-dsl · nvidia-cutlass-dsl-libs-cu13 · cuda-tile · nvidia-cudnn-frontend · nvidia-cublas-cu11 · dstack