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

2026-05-18 · Source: Hugging Face - Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

PaddleOCR 3.5, released on May 18, 2026, integrates its OCR and document parsing models with the Hugging Face Transformers ecosystem, allowing supported PaddleOCR models like PP-OCRv5 and PaddleOCR-VL 1.5 to use Transformers as an inference backend. This update introduces a flexible `engine` parameter, enabling developers to select "transformers" as the backend and configure options like `dtype`, device placement, and attention implementation via `engine_config`. While PaddleOCR continues to manage the underlying OCR and document parsing pipelines, this integration simplifies connecting these capabilities with existing PyTorch/Transformers infrastructure, particularly for applications involving RAG, Document AI, and document agents. A live demo is available on Hugging Face Spaces.

Key takeaway

For AI Engineers building RAG, Document AI, or agent applications within a Hugging Face-centered stack, PaddleOCR 3.5 significantly reduces integration friction. You can now leverage PaddleOCR's robust OCR and document parsing capabilities, such as PP-OCRv5, directly through a familiar Transformers backend, streamlining your workflow from document ingestion to LLM processing. Consider using the `paddle_static` backend if raw throughput is your primary concern.

Key insights

PaddleOCR 3.5 enables using Hugging Face Transformers as an inference backend for its OCR and document parsing models.

Principles

• Flexible inference backends enhance developer choice.
• Integration friction impedes downstream AI workflows.

Method

Install PaddleOCR 3.5, PaddleX, Transformers, and PyTorch. Initialize PaddleOCR with `engine="transformers"` and optionally configure `engine_config` for backend-specific settings like `dtype` or `attn_implementation`.

In practice

• Use `engine="transformers"` for Hugging Face-centered stacks.
• Configure `engine_config` for `dtype` (e.g., "bfloat16").
• Prioritize `paddle_static` for maximum throughput.

Topics

• PaddleOCR 3.5
• Hugging Face Transformers
• Document Parsing
• Optical Character Recognition
• Inference Backend

Code references

• PaddlePaddle/PaddleOCR

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters — PP-OCRv6 offers scalable, multilingual OCR with improved accuracy and flexible deployment options for real-world applications.
• Parse Scanned PDFs for RAG with EasyOCR: Free OCR Gives You Words, Not a Document — Traditional OCR recovers text; layout models make that text usable for RAG by adding structural context.
• Text Detection and OCR with Amazon Rekognition API — Amazon Rekognition API offers highly accurate cloud-based OCR, providing line or word-level text detection.
• DharmaOCR: Specialized Small Language Models for Structured OCR that outperform Open-Source and Commercial Baselines — Specialized small language models with DPO and SFT significantly improve structured OCR quality and stability.
• I Spent May Evaluating Different Engines for OCR — OCR is a routing problem; no single engine excels across all document types.
• Mistral OCR 3 Technical Review: SOTA Document Parsing at Commodity Pricing — Mistral OCR 3 offers superior document structure preservation and accuracy for RAG pipelines at a significantly lower cost.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Hugging Face - Blog.