Context & Challenges
As RISC-V adoption accelerates across high-performance and AI workloads, a key blocker remained: limited availability of Python packages for riscv64, especially those critical to data science and machine learning.
Within the RISE (RISC-V Software Ecosystem) initiative, the objective was to make Python a first-class citizen on RISC-V platforms, enabling developers to run modern AI and data pipelines without friction.
However, several challenges had to be addressed:
- Lack of pre-built Python wheels for riscv64 across key libraries
- Complex build and integration requirements for C/C++ and Rust-based dependencies
- Need for robust testing pipelines across diverse packages
- Performance gaps compared to established architectures
- Strict open-source licensing compliance requirements across dependencies
- Ensuring compatibility across a rapidly evolving RISC-V software stack
BayLibre contributed to accelerating the availability, performance, and reliability of Python packages on RISC-V.
Achievements
BayLibre played a key role in strengthening the Python ecosystem for riscv64, enabling broader adoption for AI and data workloads.
Key achievements included:
- Upstreamed riscv64 support for widely used Python packages such as Tornado, WebSockets, and Rust-based wheels via Maturin
- Built and maintained a growing set of Python wheels for data and ML ecosystems
- Developed a scalable vectorization library to improve performance on RISC-V platforms
- Designed and executed integration and performance testing pipelines across multiple packages
- Ensured license compliance across all builds and dependencies
- Contributed to ecosystem visibility through conference talks, technical blogs, and community engagement
These contributions significantly improved developer experience and reduced friction for running Python-based workloads on RISC-V.
Open-Source Story
This effort was deeply rooted in open-source collaboration and upstream contribution.
BayLibre worked within the RISE ecosystem to:
- Contribute patches directly to upstream Python projects
- Align with Python packaging standards (wheels, build systems, distribution flows)
- Collaborate with maintainers to ensure long-term support for riscv64
- Avoid fragmentation by ensuring changes were accepted upstream rather than maintained downstream
This approach ensured sustainability, visibility, and rapid adoption of improvements across the broader ecosystem.
Tech Stack
Customer Testimonial
“BayLibre significantly accelerated Python ecosystem readiness on RISC-V. Their work on packaging, upstreaming, and performance optimization helped close critical gaps, making riscv64 a much more viable platform for data science and AI workloads.”