Kestrel v2.0 — Zig + LuaJIT FFI inference accelerator for Gullwing. 2.97x quantized matmul, 51.6x recognition cache, hot-swappable policy layer, hybrid storage tiers, and honest benchmark discipline. Builds on colibri's foundation, contributes back what's general.
End-to-end int8 inference accelerator research stack with a custom ISA, cycle-approximate simulator, MLIR compiler, PTQ frontend, and reproducible model execution.
Add a description, image, and links to the inference-accelerator topic page so that developers can more easily learn about it.
To associate your repository with the inference-accelerator topic, visit your repo's landing page and select "manage topics."