Add Dataset Descriptions And Instructions

[rasbt]

• Adds alpaca-libre preparation script as alternative to alpaca.
• Provides instructions for all supported datasets
• Add LIMA description from Add LIMA dataset #479
• Update new token usage in LIMA
• Replace reference to --max_seq_length

[rasbt]

Let me add a documentation tutorial instead because we can download it with the existing command line args.

Will continue with the prepare_dataset.md, but just a quick check before I put more work into it: does that sound good to you @carmocca ?

[carmocca]

Yes!

[rasbt]

This should be complete now. (Or, at least good for review).

I suggest merging #466 and #447 first though, because these are the OpenWeb Text and RedPajama documents referenced at the bottom of this doc.

Note that a focus of this document is to highlight the use of --checkpoint_dir in the prepare_ scripts which a lot of people (me included) forget when trying out different models.

[rasbt]

This is the dataset tutorial companion to all the datasets we added recently (Dolly, LIMA, Alpaca Libre). It's updated after the max_token_length change and should be good to review @carmocca

[carmocca]

Thanks!