updating mnasnet docs

[jdsgomes]

Part of #5833

Adding revamp docs for MNASNET.

cc @NicolasHug

[NicolasHug]

Thanks @jdsgomes

I noticed that the 1.3 and 0.75 weights are missing / non-existent. I added them to the docs and then removed them after realizing it made the doc generation fail. Do you know if we're planning to support pre-trained weights for these models before the next release? If we don't, we might nee to change our docs to indicate that no pre-trained weights exist (and possibly remove the existence of the parameter altogether)

[jdsgomes]

Thanks @jdsgomes

I noticed that the 1.3 and 0.75 weights are missing / non-existent. I added them to the docs and then removed them after realizing it made the doc generation fail. Do you know if we're planning to support pre-trained weights for these models before the next release? If we don't, we might nee to change our docs to indicate that no pre-trained weights exist (and possibly remove the existence of the parameter altogether)

I don't know if we have plans to add those before next release, lets follow up with this offline. Merging for now, thank you for the review!

[github-actions]

Hey @jdsgomes!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py

[NicolasHug]

@jdsgomes it might still be worth updating the current docs, which for now contain ambiguous / incorrect info:

The following model builders can be used to instanciate an MNASNet model, with or
without pre-trained weights

The weights parameter is also documeted as

The pretrained weights to use

(with broken links). But if we don't support pre-trained weights, we should make sure to indicate it clearly. ATM, users will believe that we support pre-trained weights while we don't.