microsoft/hummingbird
Microsoft library that compiles trained scikit-learn, LightGBM, and other traditional ML models into PyTorch, TorchScript, ONNX, and TVM tensor computations for hardware-accelerated inference.
Velocity · 7d
+1.6
★ / day
Trend
→steady
star history
Hummingbird converts trained traditional ML models and featurizers into tensor computations executed via neural network runtimes. It bridges classical ML frameworks like scikit-learn and LightGBM with deep learning infrastructure such as PyTorch and TorchScript, enabling hardware acceleration and optimization passes that neural network frameworks provide. Users can deploy their existing models without re-engineering while gaining the performance benefits of GPU acceleration and framework-level optimizations.