Read model from cache if offline

[JosephMarinier]

Description:

Loading the sentence transformer currently fails without Internet access. Here is a simple workaround that uses sentence-transformers' own environment variables. I don't understand why it doesn't already work like that, so I opened a ticket on their end: huggingface/sentence-transformers#1725.

Checklist:

You should check all boxes before the PR is ready. If a box does not apply, check it to acknowledge it.

• ISSUE NUMBER. You linked the issue number (Ex: Resolve #XXX).
• PRE-COMMIT. You ran pre-commit on all commits, or else, you
  ran pre-commit run --all-files at the end.
• USER CHANGES. The changes are added to CHANGELOG.md and the documentation, if they impact
  our users.
• DEV CHANGES.
  • Update the documentation if this PR changes how to develop/launch on the app.
  • Update the README files and our wiki for any big design decisions, if relevant.
  • Add unit tests, docstrings, typing and comments for complex sections.

[JosephMarinier]

An alternative to checking TRANSFORMERS_OFFLINE would be to check if the local cache exists and load it if so, or even try to load the SentenceTransformer with the local cache and catch if there is an error to then load it online. I guess it depends on if it is likely to be updated online (and we want the latest) or not. Any thoughts?

[gabegma]

LGTM!!

[JosephMarinier]

This could be reverted if the issue huggingface/sentence-transformers#1725 is resolved. Let's see if the PR huggingface/sentence-transformers#1845 gets accepted.