update models docs

[ebsmothers]

A bunch of miscellaneous documentation changes to torchtune/models. Main changes:

• Provide explicit download instructions for all model variants (at least all the ones that have training configs. We can consider adding the others too, but left it out since there's not an easy one-liner to use them anyways)
• Actually document lora_dropout in builders where it's used (we should add it to the other builders where it's missing, can do that later though)
• Minor clarifications and formatting in clip_vision_encoder and TiledTokenPositionalEmbedding
• Making clip_vit_224_transform private since I don't think it's worth documenting rn (and it's not used anywhere atm)
• Improve rendering of our model-specific SentencePiece tokenizers' tokenize_messages docstrings
• Remove scale_hidden_dim_for_mlp from public APIs (similarly, I don't think it's worth documenting/exposing rn)
• Fix return type for Mistral 7B classifier

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/1167

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• [PREEMPTIVE] CPU amd64 non-GPU instances migration to Linux AMZN2023

✅ No Failures

As of commit d03cb05 with merge base c903040 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[pbontrager]

Is -hf-token <HF_TOKEN> required for phi3, gemma, and mistral? I thought it was only required for models where you had to get special approval for the weights.

[ebsmothers]

Is -hf-token <HF_TOKEN> required for phi3, gemma, and mistral? I thought it was only required for models where you had to get special approval for the weights.

Oh interesting.. I just copied from the example commands in our configs, all of which have it included. Just tested these 3 -- seems like Phi3 does not require it but the other two do. If it's all the same to you, maybe I'll leave it in for now and we can separately update the Phi3 commands across the board later on; what do you think?

[joecummings]

We need to do an overhaul of our entire models page to help people find what models we support, but this is a good improvement for our current release.