Improve discoverability of your work on HF

#2
by nielsr HF staff - opened

Hi,

Congrats on your work: https://huggingface.co/papers/2407.20445! Thank you also for sharing the artifacts (models and datasets) on the hub, and claiming the paper :)

We have some small suggestions to improve the discoverability of your work on the hub, namely:

  • the dataset: it would be great to share the datasets in separate dedicated dataset repos (which can be created at hf.co/new and then selecting a dataset repo), rather than in the model repo. The guide for uploading a dataset can be found here. It can then be pushed to the hub using push_to_hub(repo_name). For linking it to the paper, see here: https://huggingface.co/docs/hub/en/datasets-cards#linking-a-paper.
  • the model: it would be great to make download stats work for your model, along with metadata tags such that people can easily find your model. To make download stats work, we recommend leveraging the PyTorchModelHubMixin class. This class also allows users to easily specify tags which get pushed to the model card. It can be linked to the paper using the guide here.

Let us know if you need any help regarding this!

Cheers,

Niels
Open-source @ HF

Sign up or log in to comment