Burn GitHub repository (tracel Rust DL framework)

README: 'Burn is both a tensor library and a deep learning framework, optimized for numerical computing, training and inference.' Core positioning claim: training and inference usually live in separate worlds - models are trained in Python then exported to ONNX or optimized for vLLM/ONNX Runtime/TensorRT, and that export step 'is often brittle and lossy'; Burn unifies them so the exact training code runs in production. Also claims PyTorch-like ergonomics with dynamic shapes and graphs but JIT-compiles tensor-op streams with automatic kernel fusion, and incremental compilation giving sub-5-second recompiles in release mode. Topics list cuda, rocm, metal, vulkan, wasm, webgpu, onnx, kernel-fusion. MIT OR Apache-2.0. API metadata 2026-10-04: created 2022-07-18, last push 2026-10-02, 16020 stars, 1075 forks, 216 open issues, not archived, homepage https://burn.dev, no GitHub releases.