PopPy

PopPy: Automatic Parallelism Extraction for Compound AI Python Applications

PopPy is a system combining an ahead-of-time compiler with a runtime to automatically uncover and exploit parallelism in Python-based compound AI applications that invoke heavy external components such as ML models. It addresses challenges including Python language complexity, dynamic dispatch, and variable mutation while requiring minimal developer input. On real-world compound AI workloads, PopPy achieves up to 6.4× end-to-end speedup over standard Python execution while preserving sequential semantics.