FIX: Hotfix HTTP error downloading model with hub

[fepegar]

Specifying the branch when downloading a model using PyTorch Hub shouldn't be necessary, but an HTTP error was being raised anyway. There seem to have been issues with this in the past:

• Add option to skip GH validation for torch.hub pytorch/pytorch#62139
• HTTP Error 403 for torch.hub ResNet pytorch/pytorch#61755
• pretrained_resnet HTTP Error 403: rate limit exceeded n2cholas/jax-resnet#4
• Don't validate github url before downloading in hub pytorch/pytorch#61761

Explicitly specifying main seems to help, so I'm adding it to the test, and it should be included in any future documentation, just in case.

[Shruthi42]

Do we know how torch hub recommends versioning models? Do they recommend always pointing to main or to a specific release, for example?

[fepegar]

Thanks. No recommendations, really. Just possibilities. This is all they have:

github - a string with format “repo_owner/repo_name[:ref]” with an optional ref (tag or branch). If ref is not specified, the default branch is assumed to be main if it exists, and otherwise master. Example: ‘pytorch/vision:0.10’

In future docs, we can ask mention that tags should be specified for specific versions. But this is not trivial in a multi-package repo...