burn is a Rust deep learning framework that unifies training and inference through a single API where the exact code used for training runs in production. It targets developers who need to move models from experimentation to deployment without export steps.
Project overview
By running the same Rust module code for both training and production inference, burn bypasses the need for export steps between Python environments and production engines, while its sub-five-second recompile times provide a rapid feedback loop.
Project type
Model Development · Infrastructure
Deployment
Refer to project documentation
License
Apache-2.0
Best for
AI engineers and researchers comfortable with Rust who want a single codebase for training and production without export steps.
Developers targeting cross-platform on-device personalization or federated learning workflows.
Key capabilities
Executes multi-platform tensor operations via a single API where the exact code used for training is the exact code that runs in production.
Allows building burn with swappable backends, composing them to augment functionalities like autodifferentiation and automatic kernel fusion.
Transparently equips a base backend with autodifferentiation support to call backward on models.
JIT-compiles streams of tensor operations to perform automatic kernel fusion, enhancing backends with improved speed.
Allows users to follow their training progress with a terminal UI dashboard, visualizing metrics in real-time without connecting to an external application.
Limitations and risks
ONNX support currently covers a limited set of ONNX operators.
There will be breaking changes as the project is in active development.
While issues from breaking changes are likely easy to fix, there are no guarantees at this stage.
Getting started
Install Rust, add the burn crate to your Cargo.toml, define a neural network module, and run training using the Learner.
Evidence and sources
GitHub project description: Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.
README: Burn unifies the two. By executing multi-platform tensor operations via a single, unified API, the exact code used for training is the exact code that runs in production.
README: Burn is both a tensor library and a deep learning framework, optimized for
README: Burn is currently in active development, and there will be breaking changes. While any resulting issues are likely to be easy to fix, there are no guarantees at this stage.
README: most code is generic over the Backend trait, which allows us to build Burn with swappable backends.