PaddleOCR 3.5 Adds Transformers Backend for Hugging Face Integration

PaddleOCR 3.5.0, released on May 18 2026, extends the OCR library by adding a flexible inference‑engine interface that lets developers choose a backend via the engine parameter and pass backend‑specific options through engine_config. The new backend options include Transformers, which can now run PP‑OCRv5 and PaddleOCR‑VL 1.5 models directly inside a Hugging Face pipeline. Developers can set dtype, device placement, and attention implementation in engine_config, allowing the same OCR pipeline to run on Paddle static, Paddle dynamic, or Transformers backends without rewriting code. A live demo is available on Hugging Face Spaces, showcasing the float32 configuration for broad compatibility. The release keeps the core OCR and document‑parsing capabilities unchanged, but the added flexibility reduces integration friction for RAG, Document AI, and agent applications that already rely on PyTorch and Transformers. By exposing a single engine parameter, PaddleOCR now manages the internal pipeline, so users no longer need to call each component manually. The new feature is particularly useful for teams that want a familiar development experience while still leveraging the speed and tooling of the Transformers ecosystem. PaddleOCR 3.5 therefore brings OCR and document parsing closer to Hugging Face workflows while preserving the library’s original performance and extensibility.