Release RealSyn pre-trained models on Hugging Face

[NielsRogge]

Hi @Kaicheng-Yang0828 🤗

I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2502.12513.
The paper page lets people discuss about your paper and lets them find artifacts about it (your models for instance),
you can also claim the paper as yours which will show up on your public profile at HF, add Github and project page URLs.

It's fantastic to see that the RealSyn datasets (RealSyn15M, RealSyn30M, RealSyn100M) are already hosted on https://huggingface.co/datasets! This greatly improves their visibility and accessibility.

Your paper also mentions the release of "pretrained model weights," specifically the ViT-B/32 and ViT-B/16 models trained on RealSyn. Would you also be interested in hosting these pre-trained model checkpoints on https://huggingface.co/models?
Hosting them on Hugging Face will further enhance their visibility and discoverability. We can add tags to the model cards so that people can find them easily, link them directly to the paper page, and track their downloads.

If you're interested, here's a guide for uploading models. If it's a custom PyTorch model, you might find the PyTorchModelHubMixin class useful, as it adds from_pretrained and push_to_hub functionalities. Alternatively, users can download files directly using hf_hub_download.

After uploading, we can link these models to your paper page (read here) for seamless discovery. You could also consider building a demo for your models on Spaces, and we can provide a ZeroGPU grant for free A100 GPUs.

Let me know if you're interested or need any guidance :)

Kind regards,

Niels