Lit-GPT follow up

[aniketmaurya]

• Locking for multi threaded evaluation - @aniketmaurya
• implement log probs - @aniketmaurya
• make the model configuration work - @yifanmai
• add instruction for installing lit-gpt - @yifanmai

[brianwgoldman]

Not sure if this already covered, but it appears LitGPTTokenizer._tokenize_do_it also needs to be updated in order to report a 'tokens' key in the dictionary. Specifically, CachingTokenizer calls that method here and the in the next call expects the response dictionary to have an 'tokens' key here

As an alternative, it looks like LitGPTTokenizer could override _tokenization_raw_response_to_tokens.

[yifanmai]

cc @JosselinSomervilleRoberts for recent changes to LitGPTTokenizer

[yifanmai]

Closing this because we should migrate to use the new LitGPT API, and also the new HELM model_deployments.yaml format. See #1783 for details.