Output Parsers

llmabstractions.parsers.json_response.json_data_parser(llm_output: str | dict | list) → dict

Extracts dictionary or array content from an LLM response.

There are cases where you ask from the LLMs to return an array or a dictionary, but you don’t get a structured output. You can either relay on models supporting structured output, or parse the output yourself using this function.

Parameters:

llm_output (str, dict, list) – Raw LLM response. Expected as string. However, if a dictionary or list is passed, the same content will be returned as long as they can be JSON serialized.

Returns:

A dictionary or list content in the message.

Return type:

dict or list

Raises:

ParsingError: – If for any reason, the content cannot be extracted from the response.

Examples

Extract a dictionary embedded in text:

>>> llm_response = '''
... Here is the response
...
... {
...     "key1": "value1",
...     "key2": "value2"
... }
...
... Is there something else?
... '''
>>> json_data_parser(llm_response)
{'key1': 'value1', 'key2': 'value2'}