PaddleOCR 3.5: Running OCR and Document Parsing Tasks with the Transformers Backend
Released on May 18, 2026
PaddleOCR 3.5 brings OCR and document parsing tasks more closely into the Hugging Face ecosystem. With this release, supported PaddleOCR models can be run using Hugging Face Transformers as the inference backend by setting the following parameter:
engine="transformers"PaddleOCR still provides OCR model families such as PP-OCRv5, as well as document parsing model families such as PaddleOCR-VL 1.5. Transformers is now one of the supported backends for running these models.
Try the live demo on Hugging Face Spaces: https://huggingface.co/spaces/PaddlePaddle/paddleocr-3.5-transformers-demo
What’s Changed?
PaddleOCR 3.5 introduces a more flexible inference engine interface. Developers can choose a backend through the engine parameter and pass backend-specific options through engine_config.
In practice, this means:
- The pipelines behind these tasks are managed by PaddleOCR, so developers do not need to manually call each internal component.
- Transformers becomes one of the supported inference backends for running supported PaddleOCR models.
- Developers can configure backend-related options through
engine_config, such asdtype, device placement, and the attention implementation.
The overall stack can be understood simply as:
| Layer | Meaning | Example |
|---|---|---|
| Application layer | Applications that use OCR and document parsing outputs | RAG, agents, Document AI… |
| Model layer | OCR and document parsing capabilities | PP-OCRv5, PaddleOCR-VL 1.5… |
| Inference backend layer | Runtime used to run supported models | Paddle static graph, Paddle dynamic graph, Transformers |
This release mainly affects the inference backend layer: PaddleOCR continues to provide OCR and document parsing capabilities, while Transformers adds another backend option for supported PaddleOCR models, fitting naturally into Hugging Face-centric environments. Broader Document AI workflows remain under the control of developers and application builders.
Why This Matters
For RAG, Document AI, and document-agent applications, the hard part often starts before the LLM.
Developers first need to convert PDFs, scanned documents, screenshots, tables, charts, formulas, and complex page layouts into reliable structured data. If this ingestion step is not high quality, the downstream document to be continued