This is a Layer 3 domain constraint reference for building ML/AI applications in Rust. It maps ML requirements like memory efficiency and GPU utilization directly to Rust patterns, showing you why batching matters for GPUs, when to use tract versus candle versus tch-rs, and how to avoid copying giant tensors around. The code samples are practical: inference servers with OnceLock for model singletons, batched prediction pipelines, and async data loading to keep GPUs fed. It's narrow in scope but information-dense, assuming you already know the Layer 1 and Layer 2 fundamentals and just need the ML-specific constraints translated into concrete Rust decisions.
npx skills add https://github.com/actionbook/rust-skills --skill domain-ml