Google open-sources XLA Serving Engine XSE

Google has open-sourced the XLA Serving Engine, a general-purpose model export and inference runtime designed to enable a "Train in Python, Serve in C++" workflow. The C++ native runtime eliminates Python interpreter overhead, delivering significantly higher throughput and lower memory usage for non-LLM machine learning models in production environments. Core components are already deployed across Google infrastructure and are now available for external cloud TPU users and hardware partners. Open article →