Outlines v1 documentation

[cpfiffer]

Outlines v1 has changed the interface quite a bit, and so it's a good opportunity to re-write the documentation. The documentation we have at the moment works fairly well, but it could be more intentionally designed to guide the user from basics through to more advanced topics.

We should also overhaul the general theme and style. Currently, we use a hacked-together mkdocs theme that doesn't flow well visually. We should standardize on a common theme. At .TXT, we use Hextra and Hugo for our commercial documentation, and

Goals for the new documentation:

• Organized documentation flow. A common practice for documentation is the following pages:
  • Main page. Quick start, library introduction, links to various resources.
  • Getting started. A detailed guide on installation and use for our various backends.
  • Guides. Comprehensive guides to
    • Pydantic use
    • The regex DSL
    • Available inference backends
    • Using Application and Template
    • JSON schema features supported
    • Raw JSON schema development
    • Debugging + troubleshooting
    • Best practices
    • Performance tips
    • Remote serving with vLLM
    • Migration guide to v1
  • Tutorials and cookbooks. We can basically port over what we have now and change the code.
  • Advanced usage. How to use Outlines core for direct logit modification, building custom logit processors, etc.
• Move to a new URL, probably something like docs.dottxt.co/outlines/.

I'll start with generally getting the pages and general structure set up. The main page + quick start are not too hard to set up. We can mostly re-purpose what we have with some editing and polish.

[rlouf]

I think we should dedicate pages to how to use Outlines to connect to a vLLM server, an Ollama server, a llama.cpp server etc. It should be impossible to miss them when coming to the documentation.

It's a very common pattern, probably more so than using transformers, and we want to position Outlines as the SDK for structured generation.