Shardy is an MLIR-based tensor partitioning system for all dialects. Built from the collaboration of both the GSPMD and PartIR teams, it incorporates the best of both systems, and the shared experience of both teams and users.
Getting started
For model developers looking to use Shardy, take a look at Getting started with Shardy in JAX.
To start learning more about the components that make up Shardy, take a look at the Shardy Overview page.
Communication channels
- Issues and feature requests can be filed in the openxla/shardy repository.
- Questions regarding Shardy can be asked on the OpenXLA Discord.
Status
Shardy is a work in progress. The core dialect and C bindings are fully functional. The python bindings are under development.