burn
Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.
What is burn?
Burn is a next generation tensor library and Deep Learning Framework that doesn't compromise on flexibility, efficiency and portability.
Overview of burn
What it is
Burn is a tensor library and deep learning framework for numerical computing, training and inference. It unifies training and production by executing multi-platform tensor operations via a single, unified API, so the exact code used for training runs in production. Burn preserves the intuitive ergonomics of PyTorch with dynamic shapes and graphs, but JIT-compiles streams of tensor operations with automatic kernel fusion. Designed around incremental compilation, modifying model code recompiles in under 5 seconds even in release mode, delivering a Python-like feedback loop with Rust's speed and safety.
Key features
- Single, unified API for training and inference across multiple platforms.
- JIT compilation with automatic kernel fusion for performance.
- Dynamic graphs and shapes, similar to PyTorch.
- Incremental compilation for fast edit-compile-run cycles (under 5 seconds in release mode).
- Backends include CubeCL (CUDA, ROCm, Metal, Vulkan, WebGPU) and a pure-Rust Flex CPU/no_std backend.
- Part of an open-source Rust AI ecosystem with projects for model interop, domain toolkits, and benchmarking.
Use cases
- On-device personalization and federated learning.
- Teams going from prototype to deployment in a single codebase.
- Research and production workloads that require both flexibility and performance.
- Workloads that benefit from Rust's safety and speed without sacrificing a Python-like development experience.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
Licensed under Apache-2.0. Open source.
Based on the project's public README and repository metadata.