The domain-ml skill optimizes Claude's performance when assisting with Rust-based machine learning projects, focusing on the critical trade-offs between memory efficiency, GPU acceleration, and model portability. It provides structured guidance for using major Rust ML crates like Candle, Burn, tch-rs, and tract, ensuring developers implement zero-copy tensor operations, efficient batched inference, and lazy model initialization. By bridging the gap between Python-heavy research and Rust-based production environments, this skill helps users build robust, deterministic, and highly performant AI systems.
Key Features
01Standardizes model loading and portability using ONNX and the tract runtime
02Implements efficient batch processing and streaming for high-throughput data pipelines
03Provides implementation patterns for GPU-accelerated inference using Candle and tch-rs
04Optimizes memory usage through zero-copy and in-place tensor operations
05Enforces best practices for numerical precision and reproducibility in Rust ML code
06596 GitHub stars
