We (torchgeo (TorchGeo)) are working on a project that will generate 100+ pre-trained models. In the past, we’ve made a separate repository for each model, but with 100+ models we’ve started to wonder whether or not it would make more sense to stuff all of our models in a few repos instead of having 100+ separate repos. What features or functionality would we lose by doing so? Our users primarily load weights through the TorchGeo library (using timm or smp) and don’t even know that HF exists, it’s just the place we chose to distribute the files.
Late to the party, but it’s always recommended to do 1 pretrained model == 1 repo. It allows to have a download counter per model (allowing you to know which models are getting more traction), better discoverability for users on the Hub, dedicated community tabs per variant, etc.
(related: Add TorchGeo to libraries by isaaccorley · Pull Request #1464 · huggingface/huggingface.js · GitHub)
2 Likes
This topic was automatically closed 12 hours after the last reply. New replies are no longer allowed.