Add gen-like arguments to substring

[SimJeg]

Hello,

Following this issue (thanks @marcotcr for the quick reply !), would it be possible to add more gen arguments in the substring component ?

For instance name to save the result of the generation, regex to allow only substrings that respect a given pattern,max_tokens to limit the generation, stop to stop it etc.

This substring is really great because it limits hallucinations, thanks for it :)

Thanks,
Simon

[marcotcr]

Hi there. We really need to document the features in the grammar library, but in the meantime he is an example where we:

1. Save the result of a generation in a variable, using capture, even when something is optional
2. Set max_tokens only on part of the grammar, using token_limit

from guidance import capture, substring, select, regex, token_limit
# Either substring or a number with at most one token
grammar = select(
        [capture(substring('This is very cool.'), name='substring'),
         token_limit(capture(regex('\d+'), name='number'), 1)])
lm = llama2 + 'Hi there, here is a sentence "' + (grammar + '"')

lm = llama2 + 'Hi there, here is a sentence "' + (grammar + '"')
lm['substring']

Hi there, here is a sentence "This is very cool."

lm = llama2 + 'Hi there, here is a number: ' + grammar
lm['number']

Hi there, here is a number: 123