Adding architecture document

[tengomucho]

What does this PR do?

Added a document (visible here) that aims at explaining the relationship between the TGI components. This can be used to better understand TGI architecture and provide a documentation for someone who might want to implement or improve a solution deployed on a hardware different from the ones that are already supported. It can be also used as base of a reflection on choices for future design changes.

Who can review?

@OlivierDehaene @Narsil

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[drbh]

Thank you @tengomucho! LGTM

[tengomucho]

@drbh Thanks for removing the cargo lock file, I added it by mistake.
I saw the rendering is good in github.com, but not in the docs 😐 https://moon-ci-docs.huggingface.co/docs/text-generation-inference/pr_2044/en/architecture
Do you know if something can be done to fix that?

[drbh]

no worries at all, thanks for the additional docs. I can see them in production here: https://huggingface.co/docs/text-generation-inference/architecture, can you see them there as well?

I think the link shared above is from moon-ci and may just be temporary and may have been deleted with the branch

sorry I think I misunderstood your comment! I see that the mermaid docs are not rended. I think we'd need to add support for mermaid compilation (built into github) to the doc builder, and I'm not 100% how much effort that is at the moment. A short term solution is to render them to PNG for now...

ps: added a feature request here: huggingface/doc-builder#509