Help Shape Raven’s First Alpha - Call for Contributions!
I’m excited to announce that Raven’s first alpha is scheduled in 3 weeks, just in time for FunOCaml! (Following our zeroth alpha in July)
What’s New
We’ve added a plethora of new features:
- New
Talonpackage that provides an equivalent for Pandas/Polars to work with dataframes - New
Sagapackage providing tokenizers and NLP functionnalities (e.g. Ngram models) - Support for symbolic shapes and lazy views in Nx
- Support for new and machine-learning-specific data types, including boolean, bfloat16, complex16, float8, etc.
- Support for forward mode differenciation through
Rune.jvp - Support for automatic vectorization through
Rune.vmap - Addition of a complete linear algebra suite in Nx
And, of course, since it is the theme of the FunOCaml workshop: addition of transformers blocks in Kaun to train large language models.
All of these features are currently in-progress in Raven’s repository, ranging from ready-to-release (e.g. Talon) to prototyping (e.g. Kaun’s transformers), but we’re aiming for a release with all of the above mid-september.
Call for Contributions
If you were looking to contribute to Raven, now’s probably a really good time! There’s going to be a lot of activity in the coming weeks, and the repository is in good shape for first-time-contributors, with a lot of good first issues: we’ve just landed a complete linear algebra suite that’s ~80% working, which makes for the perfect set of starter issues: failing tests that need fixing.
- 15 failing tests on the native backend
- 35 on the C backend
These range from implementing frontend operations using our new backend ops, to investigating bugs in the newly added linear algebra functions. Pick a test, fix it, open a PR!
Getting Started
git clone https://github.com/raven-ml/raven
dune pkg lock
dune build @nx/runtest
If you’re interested in contributing, don’t hesitate to reach out as well,
I’m happy to give pointers and show you around the codebase.
Hoping you’re as excited about the upcoming release as I am - Happy Hacking!