An illustration of a brown and a white hand coming towards each other to pass a business card with the logo of Quansight Labs
Back to blog

A second CZI grant for NumPy and OpenBLAS

Published November 19, 2020



Melissa Mendonça

I am happy to announce that NumPy and OpenBLAS have once again been awarded a grant from the Chan Zuckerberg Initiative through Cycle 3 of the Essential Open Source Software for Science (EOSS) program. This new grant totaling $140,000 will fund part of our efforts to improve usability and sustainability in both projects and is excellent news for the scientific computing community, which will certainly benefit from this work downstream.

Why this is important for NumPy and OpenBLAS

The Essential Open Source Software for Science program is visionary and unique. The activities supported by this funding are not restricted to new technical features. Rather, it supports maintenance, community engagement, and other types of work that often go unrewarded in open source projects. And although its primary goal is to fund the computational foundations of biology, the benefits of the work done under the Chan Zuckerberg Initiative's program extend well beyond specific scientific areas.

This is the second grant from CZI that NumPy and OpenBLAS have received. This past year, the funding has enabled several high-impact improvements to the NumPy and OpenBLAS communities. For NumPy, this has meant a new website, the establishment of the Documentation and Website Teams, and mentorship and onboarding activities to grow and diversify the maintainers for the project. For OpenBLAS, the focus has been on technical improvements (thread safety, support for the 512-bit Advanced Vector Extensions (AVX-512) code, and thread-local storage). The impacts of these developments extend throughout the scientific Python ecosystem and beyond, and the increased activity has also attracted several new contributors.

Our plans for 2021 include continuing with our efforts in documentation and community building for NumPy and also working to modernize its integration with Fortran tools via F2py, in addition to dedicating resources towards ensuring the sustainability of both NumPy and OpenBLAS. You can read the full proposal here.


The establishment of the NumPy documentation team has resulted not only in a restructuring of the documentation and a new repository for educational content but has also created the opportunity for many new contributors to join the project. Part of this new grant will enable us to continue that work, resulting in a set of high-quality documents around NumPy while at the same time diversifying opportunities for community contributions.


As a part of NumPy, the F2py tool is used by numerous projects in the Python scientific ecosystem, including SciPy itself, to automate the compilation of Fortran code into Python modules, making it an essential piece of the infrastructure for these projects.

However, Fortran has evolved over the years. Although considered by some as a niche language, it is still very popular among scientific computing and high-performance developers and is in active development. Due to a lack of dedicated development work in the past years, F2py does not support many current features of the language. In addition, because of limited resources dedicated to testing and maintenance, regressions have been introduced in the F2py code that may result in a propagation of bugs for many other projects. With this grant, we plan to address the maintenance backlog and extend the functionality of F2py and bring back Pearu Peterson, the original author for the F2py tool, to its development.

Currently, F2py supports wrapping Fortran 77 and Fortran 90 code, except for derived types introduced in Fortran 90. To support modern use cases, we plan to add support for derived types and work towards full Fortran 2003 compatibility. While there have been attempts at resolving these issues in the past, they were never integrated into NumPy, and at the moment, this would require a larger refactor of F2py, which we propose to do. We are also writing a new F2py user guide with comprehensive examples and feature descriptions.


OpenBLAS is a highly optimized library for linear algebra computations. It provides accelerated versions of BLAS (Basic Linear Algebra Subroutines and LAPACK (Linear Algebra Package), which are the two universal interfaces to perform linear algebra with. It is a key dependency of NumPy and SciPy, as well as of R and Julia. While most users will not realize they are using it, it is essential for the three leading scientific open-source programming ecosystems.

This grant will enable the OpenBLAS maintainers to continue addressing critical technical issues. These include issues with Thread Local Storage, performance at certain problem sizes, and the implementation of BLAS extensions already provided by similar libraries.

In practice

As part of the third cycle of the ​Essential Open Source Software for Science (EOSS)​ program, CZI awarded $3 million for 17 new grants. These awards bring the total number of funded proposals to 67 projects and the EOSS program's total commitment to funding scientific open source to $11.8 million. View the ​full list of grantees​.

As for our project, the funds will be again managed by NumFOCUS as a fiscal sponsor. Of the total value (removing the overhead charges), about two-thirds will go to NumPy. This will support me, as the PI of this proposal, and Pearu Peterson part-time for a year, as a subcontract with Quansight. The remaining funds will support Martin Kroeker part-time for a year.

To work!

The actual work begins on January 1st, and I'm looking forward to this. Moving from being a part of the grant project for NumPy in 2020 to being the PI in 2021 is exciting - I can't wait to see what we do next!

More articles from our Blog

An illustration of three cartoon characters - CuPy Cube, SciPy Snake and PyTorch Flame - together (happily!) under a tree with green leaves bearing the Consortium for Python Data API Standards logo ("DataAPIs") on a brown background. CuPy Cube and PyTorch Flame are made up of the CuPy and PyTorch logos respectively, with smiley faces and stick legs. SciPy Snake is a green snake with a red tongue, wearing sunglasses with SciPy logos for lenses.

The Array API Standard in SciPy

By Lucas Colley

October 4, 2023

The Data APIs logo next to the scikit-learn logo.

Array API Support in scikit-learn

By Thomas J. Fan

September 19, 2023