Adding LightGlue entry

[ariG23498]

The root issue for context: alexstoken/image-matching-models#45

I am trying to add more custom image matching libraries to the hub, so that we can track their download better and have more models uploaded to the Hub.

CC: @pcuenca

[Vaibhavs10]

are these the only models affected by it? https://huggingface.co/models?other=lightglue

(should there be more?)

[Wauplin]

Wondering if the download count rule is necessary. The rest looks good 👍

[Wauplin]

Looking at the 5 repos in https://huggingface.co/models?other=lightglue, they all have the same structure with config.json + model.safetensors but no .pth file. Since downloads are counted by default based on config.json, the download counter is already enabled for these models (see screenshot). Is there a reason for changing that?

[Wauplin]

Note that it's still possible to register lightblue as a library for better metadata + link to GH repo but I'd remove the download counter rule to fallback to default one:

Suggested change

countDownloads: `path_extension:"pth"`,

[ariG23498]

I see that lightglue is part of transformers which helps in building the config.json when uploaded. (@Vaibhavs10 these are the models that are uploaded to the Hub at the moment)

What I wanted to do was to upload all the checkpoints from the GitHub release to the Hub and try to track downloads from it.

@Wauplin I agree that .pth as a counter would be counter intuitive, but for the sake of making least code changes and also helping people track number of downloads, I would need to track the .pth files. What do you suggest I do here?

[pcuenca]

The counting rule is there because the original library code (here and through this popular wrapper) doesn't use the transformers implementation or a config.json file.

I would suggest to add a download counting rule that covers both cases:

		countDownloads: `path_extension:"pth" OR path:"config.json"`,

@ariG23498, if we only use path_extension:"pth" then we'd disable the counts for the other models.

[Wauplin]

Agree with @pcuenca on how to proceed!