Improve HF integration

[NielsRogge]

Hi,

Niels here from the open-source team at Hugging Face. I discovered your work through the paper page: https://huggingface.co/papers/2408.15980.

Thanks for making the models and dataset available on the 🤗 hub. Some small suggestions on how to improve the discoverability of your work:

• we can add a "robotics" tag to your model repository, so that people can discover it from https://huggingface.co/models?pipeline_tag=robotics.
• we can link the model and dataset repositories to the paper page, as explained here: https://huggingface.co/docs/hub/en/datasets-cards#linking-a-paper.
• usually we recommend to push separate checkpoints to separate model repositories (for things like download stats to work). In this case, you could for instance have a mlfu7/crossmae-vit-base model repository (with tags like "image-classification, "robotics" etc.), as well as a mlfu7/llama-7b-lora model repository, etc. See here for a guide: https://huggingface.co/docs/hub/models-uploading. For custom PyTorch models, the easiest is to leverage the PyTorchModelHubMixin class.
• the various model repos (as well as the dataset, paper) can then be grouped together into a collection. See e.g. here for current trending collections: https://huggingface.co/collections.

Let me know if you need any help regarding this!

Cheers,

Niels
ML Engineer @ HF 🤗

[Max-Fu]

Thanks for the suggestions! We will integrate them in the next two weeks!