PaddleX

PaddleX 3.0 is a low-code development tool built on the PaddlePaddle framework and published by the PaddlePaddle organization. It integrates pretrained models and supports end-to-end development from training to inference across several mainstream hardware platforms. The project is open source under the Apache 2.0 license, and its documentation lists v3.7.2 as the current release.

What It Is

PaddleX packages models into task-oriented pipelines, meaning predefined workflows that combine one or more single-function modules. According to the README, it covers 200+ PaddlePaddle models organized into 33 pipelines and 39 single-function modules. Areas include text-image analysis, OCR, object detection, time-series forecasting, speech recognition, video analysis and document-oriented vision-language models.

Workflow and Usage

Users install PaddlePaddle and then paddlex[base] with pip, which requires Python 3.8 to 3.13. A pipeline can be run from the command line with paddlex --pipeline <name> --input <file> --device <device>. In Python, create_pipeline() and predict() return results that can be printed or saved as images, JSON, and, for table pipelines, XLSX or HTML. Pipelines can be fine-tuned with the secondary development capability when pretrained results are not sufficient.

Deployment and Hardware

The docs describe high-performance inference, serving deployment, on-device deployment, and ONNX model export. Supported devices include CPU, GPU, Kunlunxin XPU, Ascend NPU, Cambricon MLU and Hygon DCU, selected through the device parameter.

Recent Updates

The README lists v3.3.0 (October 2025) adding inference and deployment support for PaddleOCR-VL and PP-OCRv5 multilingual models. v3.2.0 added CUDA 12 support for high-performance inference and open-sourced the serving solution. v3.1.0 added the PP-DocTranslation pipeline, which uses PP-StructureV3 and ERNIE 4.5 Turbo.