PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend

PaddleOCR 3.5 has been released, bringing optical character recognition (OCR) and document parsing tasks closer to the Hugging Face ecosystem. This latest version allows supported PaddleOCR models to run with Hugging Face Transformers as an inference backend, simply by setting a configuration option. PaddleOCR continues to provide a range of OCR model series, including PP-OCRv5, and document parsing model series, such as PaddleOCR-VL 1.5.

With the release of PaddleOCR 3.5, Transformers becomes one of the supported backends for running these models. Developers can try out a live demo on Hugging Face Spaces: https://huggingface.co/spaces/PaddlePaddle/paddleocr-3.5-transformers-demo. The new release introduces a more flexible inference-engine interface.

Developers can select the backend through the engine parameter and pass backend-specific options through engine_config. This allows for greater flexibility and customization when running OCR and document parsing tasks. According to the developers, the hard part of building Document AI workflows often starts before the large language model (LLM) stage.

Developers first need to turn PDFs, scanned documents, screenshots, tables, charts, formulas, and complex page layouts into reliable structured data. PaddleOCR helps address this document ingestion challenge by providing OCR series models, such as PP-OCRv5, and document parsing series models, like PaddleOCR-VL-1.5. With PaddleOCR 3.5, these capabilities are now easier to connect with Transformers-centered stacks.

Supported PaddleOCR models can run with a Transformers backend, while PaddleOCR continues to manage the OCR or document parsing pipeline behind the scenes. This means less integration friction and a more natural path from documents to downstream RAG, agent, search, analytics, or automation workflows for developers. The PaddleOCR team would like to thank the Hugging Face engineers who supported the PaddleOCR 3.5 Transformers integration, particularly Anton Vlasjuk, Raushan Turganbay, and Yoni Gozlan, for their valuable contributions and feedback.