The Flower team is excited to announce the release of Flower 1.31 stable, packed with new updates!
Flower is a friendly framework for collaborative AI and data science. It makes novel approaches such as federated learning, federated evaluation, federated analytics, and fleet learning accessible to a wide audience of researchers and engineers.
Thanks to our contributors
We would like to give our special thanks to all the contributors who made the new version of Flower possible (in git shortlog order):
Charles Beauville, Chong Shen Ng, Daniel J. Beutel, Daniel Nata Nugraha, Gustavo Bertoli, Heng Pan, Javier, Leandro, Lorenzo Sani, Micah Sheller, Mohammad Naseri, Nyam2C, Patrick Foley, Seulki Yun, Sijiaomg Ohoh, Taner Topal, Yan Gao, Young D. Kwon, reducedradius
What's new?
-
Prepare support for Flower Agent (#7215, #7216, #7220, #7238, #7252, #7253, #7256, #7261, #7264, #7272, #7273, #7277, #7287, #7291, #7292, #7293, #7301, #7314, #7324, #7337)
Prepares the foundation for running Flower Agent on SuperGrid, introducing the infrastructure and building blocks needed for upcoming agentic AI workflows.
-
Add SuperGrid guides (#7188, #7221, #7257, #7279, #7328, #7334)
Adds new and expanded SuperGrid guides, including Flower CLI usage and instructions for running apps on SuperGrid. To learn more about using SuperGrid, see the Get started with Flower tutorial.
-
Add Executor Abstraction for SuperExecs (#7174, #7206, #7231, #7232, #7233, #7234, #7235, #7323)
Introduces a SuperExec Executor framework that separates task selection from how tasks are launched. The new execution specification, launch result handling, token-file support, and capacity hook make it easier to add and enrich launch paths for different execution environments, from local subprocesses to containers, Kubernetes pods, and other managed runtimes.
-
Improve high availability (#7169, #7173, #7265, #7266, #7282, #7284, #7285, #7303)
Improves high availability by making cleanup and startup operations safer, and strengthening concurrent node, message, and run handling.
-
Track ClientApp runtime metrics in simulation runs (#7248, #7250, #7251, #7260)
Tracks ClientApp runtime metrics in simulation runs by adding run metrics to the AppIo proto, collecting run metrics, and forwarding simulation run metrics.
-
Add task-to-task communication (#7156, #7178, #7194, #7213)
Adds the base structure for task-process-to-task-process communication on the same host.
-
Restructure framework modules (#7242, #7246, #7259, #7278)
Restructures flwr by relocating Context, Message, and RecordDict to flwr.app, and Grid and ServerApp to flwr.serverapp.
-
Migrate framework CI/CD to uv (#6693, #7202, #7214, #7223, #7275)
Migrates framework CI/CD to uv, including the Poetry replacement, uv-build backend switch, and CI workflow update.
-
Update translations (#7222, #7227, #7228, #7230, #7245, #7255, #7270, #7286, #7289, #7295, #7296, #7302, #7309)
-
Upgrade Ray to 2.55.1 (#7181)
-
General improvements (#7155, #7218, #7219, #7229, #7243, #7254, #7262, #7267, #7281, #7306, #7315, #7331)
As always, many parts of the Flower framework and quality infrastructure were improved and updated.
Incompatible changes
-
Drop Python 3.10 support (#7190)
With Python 3.10 reaching end-of-life in October 2026, Flower drops support for Python 3.10.