Rust developers wanting to run or deploy ML models without a Python runtime dependency.

Candle is a Hugging Face maintained minimalist machine learning framework for Rust with CPU and CUDA GPU backends and WebAssembly support.

Used as a Rust crate embedded directly in applications and servers, or compiled to WebAssembly for browser deployment.

Released by Hugging Face around 2023 and actively maintained as of 2026.

Enables fast, memory safe, dependency light model inference for LLMs and diffusion models in performance sensitive or embedded Rust environments.

Distributed as a Rust crate via crates.io, with example model implementations for LLaMA, Whisper, and Stable Diffusion in the same repository.

Open source under a dual Apache-2.0 and MIT license, maintained by Hugging Face; runs fully locally or in browser with no required external API calls.