Show HN: TiGrIS, a tiling compiler that fits ML models onto embedded devices

Category: infrastructure

Tags: tiling-compiler, ml-embedded, onnx

Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 7, Utility: 7)

TiGrIS is an ahead-of-time compiler that tiles and partitions ML models to fit into tight SRAM budgets on embedded devices, emitting binary plans executed by a zero-dynamic-allocation runtime. It offers a novel approach by rearranging computation rather than shrinking the model, allowing powerful models to run on hardware with limited memory. The project is interesting for its practical combination of compiler techniques and embedded constraints, with clear potential to bridge the gap between ML capability and edge devices.

Target audience: embedded-devs, ml-engineers

Repository: https://github.com/raws-labs/tigris · Python · Apache-2.0

View on Hacker News