Skip to content

v0.3.0a1

Pre-release
Pre-release

Choose a tag to compare

@tongzhou80 tongzhou80 released this 28 Nov 08:08
· 190 commits to main since this release

Added

  • True just-in-time compilation: compilation now occurs at loop execution time instead of function invocation, allowing the compiler to utilize full runtime context.
  • Automatic device detection and data movement: runtime automatically selects the appropriate device and moves data as needed.
  • Automatic reduction detection: identifies reduction operations for scalars and constant-indexing arrays.
  • Initial multi-backend code structure: lays the groundwork for supporting multiple backends (Triton, CUDA, etc.).
  • GitHub CI/CD integration: continuous integration and testing added for improved code quality.
  • Additional testing: new tests added to cover core functionality and compiler workflows.

Changed

  • Refactored JIT and internal API to support runtime compilation and multi-backend extensibility.

Notes

  • This is an alpha release; some features are experimental and APIs may change in future versions.
  • Feedback from early users is encouraged to help stabilize the final 0.3.0 release.