Suggestion on how to improve your code using json_repair

[mangiucugna]

Hi, I am the maintainer of json_repair and noticed that you can refactor this code and make it faster: https://github.com/ju-bezdek/langchain-decorators/blob/539bec41bffdf3a893ba0762788326dee2d7c21f/src/langchain_decorators/output_parsers.py#L133C5-L145C5 like so:

from json_repair import loads
try:
  json_dict = loads(json_str)
except ValueError
  raise OutputParserExceptionWithOriginal(msg, text, error_code=ErrorCodes.INVALID_JSON)

this version makes your code dependent on json_repair so you might not want that, in that case you still use the options return_objects=True and skip_json_loads=True to make your code faster since those will avoid the additional json.loads() and return an object and not a string (as you want in your code)

Cheers

[ju-bezdek]

Thanks for the feedback !

I'll add it to the next release

... yeah, i didnt want the dependency, since its really rare these days that LLM wont produce valid JSON, (depends on the model ofc) ... but it still happens.

[mangiucugna]

great to hear, if you encounter any bug or have any feature request just open an issue!