gpt-utils

Helper tools for use with OpenAI's GPT-3 API.

The API key can be provided in three ways:

• In the environment variable OPENAI_API_KEY
• Via class attribute GPT.api_key of class GPT in module gpt_utils.core
• Via keyword argument api_key in any method or function which obtains a response from OpenAI's servers

Disclaimer

These don't work perfectly.

gpt_utils.utils

from gpt_utils import utils

utils contains simple functions which transform text in a specific way. Some functions are bidirectional; these take an optional reverse keyword argument which reverses the conversion's input and output.

utils.case

>>> from gpt_utils.utils import case

Contains functions which modify the case of words in a string. These functions are meant to selectively preserve the case of proper nouns, etc. with minimal involvement. The first, title(), takes a str containing a word or phrase with proper capitalization for use in a heading or title:

>>> case.title('the ant: an introduction')
'The Ant: An Introduction'

The second, mid_sentence(), adjusts the case of input words/phrases for insertion mid-sentence:

>>> case.mid_sentence('bicycles on inman street')
'bicycles on Inman Street'
>>> case.mid_sentence('An italian restaurant in somerville')
'an Italian restaurant in Somerville'

utils.location

>>> from gpt_utils.utils import location

Currently contains one function, fix_location(), which takes an unformatted string containing a location and formats it properly for display:

>>> location.fix_location('lenox ma')
'Lenox, MA'

utils.number

>>> from gpt_utils.utils import number

Contains two functions which convert numbers to strings. The first, number_to_str(), takes an int or float and writes out the same number as a str:

>>> number.number_to_str(83)
'eighty-three'
>>> number.number_to_string(3.14)
'three point one four'

The second, number_to_ordinal(), takes an int and returns it in ordinal form as a str:

>>> number.number_to_ordinal(43)
'forty-third'

utils.dechatify()

I have found that GPT tends to produce output of lower quality when it is prompted with text containing Internet/SMS vernacular. As a remedy, dechatify() converts text to standard/written English:

>>> utils.dechatify('how r u')
'How are you?'
>>> utils.dechatify('if the implementation is easy to explain then it might b a good idea')
'If the implementation is easy to explain, then it might be a good idea.'

utils.generate_stem()

This function accepts a str containing a word/phrase and returns a sentence stem pairing the word/phrase with the appropriate form of the verb "to be":

>>> utils.generate_stem('GitHub')
'GitHub is'
>>> utils.generate_stem("pleiades")
'The Pleiades are'