parseq and vitstr_base model on PyTorch: getting "Invalid model URL, using default initialization error" on 0.7.0

[dchaplinsky]

Bug description

I'm using the latest public release and trying to load vitstr_base or parseq models on PyTorch.

While release notes clearly states Two new recognition models was added (ViTSTR and PARSeq) in both frameworks by @felixdittrich92 @nikokks seems that those will be added in 0.7.1. Any plans to release those, or should I use pip install from GitHub for now?

Code snippet to reproduce the bug

self.model = ocr_predictor(
            det_arch='db_resnet50',
            reco_arch='parseq',
            pretrained=True
        )

### Error traceback

Invalid model URL, using default initialization.

### Environment

python-doctr==0.7.0


### Deep Learning backend

Collecting environment information...

DocTR version: v0.7.0
TensorFlow version: N/A
PyTorch version: 2.1.1+cu121 (torchvision 0.16.1+cu121)
OpenCV version: 4.8.1
OS: Ubuntu 23.10
Python version: 3.9.18
Is CUDA available (TensorFlow): N/A
Is CUDA available (PyTorch): Yes
CUDA runtime version: Could not collect
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 525.147.05
cuDNN version: Could not collect

[felixdittrich92]

Hey @dchaplinsky 👋,

That's correct the pytorch models are only available at main branch (v0.8.0a) atm and will be fully published with the next release (planned in january) :)

[felixdittrich92]

Feel free to test from main branch 🤗

[dchaplinsky]

@felixdittrich92 no updates on 0.8.0 release yet?

[felixdittrich92]

Hey @dchaplinsky 👋

Still waiting that some runs finish :)
If #1364 is finished we will do the release