On this week's AI board

shoehorn: quantize a model to fit your exact memory

Rust tool that solves a per-tensor mixed-precision quantization so a model fills the memory you actually have.