Show HN: An experimental tensor DSL/compiler with explicit scheduling

Category: compiler

Tags: tensor-compiler, dsl, rust, flash-attention, scheduling

Score: 6.0/10 (Innovation: 7, Technical: 7, Documentation: 7, Utility: 3)

This project is an experimental tensor DSL/compiler written in Rust that introduces explicit scheduling at the language level to enable predictable loop fusion, storage folding, and online reduction rewriting. It demonstrates a novel approach to implementing FlashAttention with a minimal and principled scheduling model, though it remains a proof-of-concept with limited real-world utility.

Target audience: backend-devs

Repository: https://github.com/ilang-dev/i · Rust · 1 stars

View on Hacker News