[Feature]: GBNF Support

[Autumnlight02]

🚀 The feature, motivation and pitch

To align the feature set of llamacpp and Aphrodite

Alternatives

Grammar using the openai spec, but thats nowhere close.

Additional context

https://github.com/ggerganov/llama.cpp/tree/master/grammars
May be interesting for mass usage, and I would personally need it for some game projects which always need a structured output but also the throughput only aphrodite can offer

[sgsdxzy]

I think aphro already supports EBNF, and various guided decoding options provided by outlines/lm-format-enforcer. What feature is missing from outlines but present in llamacpp?

[Autumnlight02]

Wiki has no results for EBNF

[sgsdxzy]

There are various guided_* options in the api, if you want context-free grammars you should be using guided_grammar
https://github.com/dottxt-ai/outlines?tab=readme-ov-file#using-context-free-grammars-to-guide-generation is the documentation for outlines.
But guided_json or guided_regex should have covered most use cases.