PaddlePaddle/PaddleX

Provides unified Python APIs and model pipelines for 270+ pretrained models across vision, OCR, time-series, and document understanding tasks, packaged into 33 production-ready pipelines. Supports full-stack development from training through high-performance inference, ONNX export, and multi-hardware deployment (NVIDIA, Kunlun, Ascend, Cambricon) via CLI, SDK, or cloud-based GUI. Enables complex multi-model compositions—like PP-OCRv5's 15-model chain and PP-StructureV3's document parsing—through standardized inference interfaces and pre-configured workflows.

6,082 stars. Actively maintained with 24 commits in the last 30 days.