hummingbird

Why we included this project

Hummingbird takes trained classical models, the kind you build with scikit-learn, LightGBM, or XGBoost, and rewrites them as tensor computations that run on PyTorch, TorchScript, ONNX, or TVM. That opens up hardware acceleration and whatever optimizations those frameworks add later, without asking you to retrain or re-engineer anything. The inference API follows scikit-learn's shape, so a compiled model often drops into existing serving code unchanged. For teams that would rather support one deployment stack for both tree-based and neural models, or that need more throughput from mature gradient-boosted pipelines, this is a low-friction way to get it.