---
title: "PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend"
description: "PaddleOCR 3.5 adds a more flexible inference-engine interface for OCR and document parsing, allowing supported models to run with a Hugging Face Transformers backend selected through engine=\"transformers\". It keeps pipelines managed by PaddleOCR while exposing backend-specific settings such as dtype, device placement, and attention implementation through engine_config. The release supports command-line and Python usage, with examples configuring GPU execution and float32, bfloat16, or SDPA options according to hardware and deployment environment. This provides a more natural path for RAG, Document AI, search, analytics, and agent stacks centered on PyTorch and Transformers, while the default paddle_static backend is usually recommended when maximizing OCR or document parsing throughput is the priority."
---

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

[Hugging Face](https://yomu.fyi/company/hugging-face) · AlexZhang, cuicheng, Jun Zhang, Manhui Lin, Yue Zhang · May 18, 2026

**Type:** Announcement

## Summary

PaddleOCR 3.5 adds a more flexible inference-engine interface for OCR and document parsing, allowing supported models to run with a Hugging Face Transformers backend selected through engine="transformers". It keeps pipelines managed by PaddleOCR while exposing backend-specific settings such as dtype, device placement, and attention implementation through engine\_config. The release supports command-line and Python usage, with examples configuring GPU execution and float32, bfloat16, or SDPA options according to hardware and deployment environment. This provides a more natural path for RAG, Document AI, search, analytics, and agent stacks centered on PyTorch and Transformers, while the default paddle\_static backend is usually recommended when maximizing OCR or document parsing throughput is the priority.

## Context

Document ingestion often precedes RAG, Document AI, search, analytics, and agent workflows. PDFs, scanned documents, screenshots, tables, charts, formulas, and complex layouts must be converted into reliable structured data; weak ingestion can cause downstream workflows to miss information, retrieve incorrect context, or produce unreliable answers.

## Approach / What changed

PaddleOCR 3.5 introduces an inference-engine interface that selects the backend with engine and passes backend-specific settings through engine\_config. Supported PaddleOCR models can run through Transformers while PaddleOCR continues managing the OCR or document parsing pipelines. Configuration examples include dtype, device type and ID, and attention implementation, with command-line and Python APIs provided.

## Takeaways

- The Transformers backend is selected with engine="transformers", while engine\_config supports options including dtype, device placement, and attention implementation.
- PaddleOCR continues to manage the pipelines behind supported OCR and document parsing tasks, so developers do not need to call each internal component manually.
- The Transformers backend suits Hugging Face- and PyTorch-centered integration, but paddle\_static is usually recommended when maximum OCR or document parsing throughput is the priority.

**Tags:** [Deployment](https://yomu.fyi/topic/deployment), [Developer Experience](https://yomu.fyi/topic/developer-experience), [Performance](https://yomu.fyi/topic/performance)

- Source: [Hugging Face](https://huggingface.co/blog/PaddlePaddle/paddleocr-transformers)
- Source URL: https://huggingface.co/blog/PaddlePaddle/paddleocr-transformers
- Ingested by Yomu: 2026-08-29T01:49:04.493Z

[Read original post](https://huggingface.co/blog/PaddlePaddle/paddleocr-transformers)