PaddleOCR 3.5 增加 Transformers 後端支援
paddleocr 3.5 為其 pp-ocrv5 和 paddleocr-vl 1.5 模型新增了 transformers 後端，以簡化與 hugging face 工作流程的整合。
paddlepaddle released paddleocr 3.5 with a flexible inference-engine interface that adds transformers as a supported backend alongside the default paddle_static. developers can now run pp-ocrv5 and paddleocr-vl 1.5 models within transformers-centric stacks, simplifying the ingestion of complex documents for rag and document ai applications. the release manages pipelines internally while allowing specific backend configurations such as attention implementation and device placement. this interoperability reduces integration friction for teams already using pytorch and transformers.