Parse output as a dictionary object

When prompting an LLM for structured (e.g. return JSON), you might get free text in addition to the requested content, or partially structured content. To convert the LLM output to a structured format, you can either relay on the LLM provider to do the parsing for you, or parse it inside your codebase.

The json_data_parser function, will try to extract dictitonary or array content from an LLM response and raise an exception when it fails.

from llmabstractions import json_data_parser

llm_response = '''
Here is the response
```json
{
     "key1": "value1",
     "key2": "value2"
 }
```

Is there something else?
'''
parsed_response = json_data_parser(llm_response)
parsed_response

{'key1': 'value1', 'key2': 'value2'}

In other cases, the content might be embeded in the text response. E.g.

llm_response = '''
Here is your array

[
    {
        "key1": "value1"
    },
    {
        "key2": "value2"
    }
]

Is there something else?
'''
parsed_response = json_data_parser(llm_response)
parsed_response

[{'key1': 'value1'}, {'key2': 'value2'}]

The json_data_parser function, will raise a ParsingError when it fails to extract JSON content from the text.

llm_response = '''
Here is your array

There is nothing to extract here
'''
parsed_response = json_data_parser(llm_response)
parsed_response

---------------------------------------------------------------------------
ParsingError                              Traceback (most recent call last)
Cell In[6], line 6
      1 llm_response = '''
      2 Here is your array
      3
      4 There is nothing to extract here
      5 '''
----> 6 parsed_response = json_data_parser(llm_response)
      7 parsed_response

File ~\dialectos-ai\repos\llmabstractions\llmabstractions\parsers\json_response.py:71, in json_data_parser(llm_output)
     68             parsed_response = _extract_dict_or_array_from_text(llm_output)
     70 if parsed_response is None:
---> 71     raise ParsingError('Unable to parse LLM output.')
     73 return parsed_response

ParsingError: Unable to parse LLM output.