Seeking contribs for Ez: Nx CPU bindings without jvm/bazel/python toolchain with zig instead

Seeking contributors for Ez. Sadly I don’t have enough time to get this up and running.

It’s a backend for Nx (Jose’s numerical computation library) that ditches jvm/bazel/python and instead uses Zig as a compiler toolchain to generate tensor operations, the code heavily uses elixir and Zig metaprogramming to use composition behind the scenes and keep the code legible.

Currently I have broadcasted tensor addition and multiplication (across all native types - i8, i16, i32 i64, u8, u16, u32, u64, f32, f64) up and running, which should give contributors an idea of how to extend this across all of Nx’s supported operations.

12 Likes