More documentation for types, examples with models without tokenizers

[handrew]

Hey there, I think I came across this repo because GitHub Explore suggested it to me. Thanks for putting this out. I do a lot of prompt engineering to get LLMs to output clean JSON files and it is maddening how often the data returned will be malformatted in some trivially small way, so it's great to see something like this.

Had two requests for you, time permitting.

1. Could you add more documentation about what types are supported? I see from the example in the README that there's strings, numbers, and booleans. Anything else? Differentiation between floats and ints?
2. I see that the API interface requires a tokenizer. Does this mean this package can only be used for Hugging Face models? I'd love to use OpenAI's GPT3.5/4 with it if possible.

Thanks again, really appreciate your work.

[MHillier98]

Seconding this, I was trying to see whether I could get this to return an array of objects but haven't been able to figure out the syntax for that, or if its even possible