Candle
Hugging Face's minimalist Rust tensor library, Metal backend included.
When to use it
Use this when you're shipping a Rust binary and want a small, dependency-light inference engine with Metal support rather than pulling in Python.
Alternatives in Runtimes
MLX
Apple's own array framework, built so unified memory stops being a compromise.
MLX LM
The package behind every `mlx_lm.generate` command on this site.
MLX Swift
MLX's tensor core, ported so Swift apps don't need a Python bridge.
llama.cpp
The C++ runtime that made GGUF and Metal inference normal.
MLC LLM
A compiler-first runtime that targets Metal, CUDA and mobile from one graph.