Can I run paddle with models downloaded locally?

[MarekHerty]

Hi,
I got really excited when the new verion coma out, However when I downloaded the model inference files it does not seem to work, It wants to dowload the model which is not possible because the server I am using has no access to internet,

ocr = PaddleOCR(
    use_doc_orientation_classify=False,
    text_detection_model_name="PP-OCRv5_server_det",
    text_detection_model_dir="./PP-OCRv5_server_det_infer",
    text_recognition_model_name="PP-OCRv5_server_rec",
    text_recognition_model_dir="./PP-OCRv5_server_rec_infer",
    use_doc_unwarping=False,
    use_textline_orientation=False,
)

=> RuntimeError: Download from https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_det_infer.tar failed. Retry limit reached

The files that get dowloaded in model.tar are:
inference.json
inference.pdiparams
inference.yml

Any idea what am I doing wrong? or what steps to take to fix this? In addition as I understand it the only models that support latin are v3 models (However I cant seem to find the recognition models, have they been retired?)

[liuhongen1234567]

Hello, there is no issue with the model you downloaded; the JSON format is the default new model format for Paddle 3.0.
You can find the Latin model by clicking on 👉Details of the Model List in the text recognition module in PPOCRv5 Documentation.

[cuicheng01]

Thank you for raising this question. We will soon host all PaddleOCR 3.0 models on Hugging Face, providing multiple download sources. By default, the models will be downloaded from Hugging Face.