API Reference

Utils

Query Decomposition

The QueryDecomposition class is used to decompose a complex question into multiple simpler questions using a LLM model. The class uses a default decomposition prompt from the Visconde paper. The default prompt is based on few-shot prompts from the StrategyQA dataset.

Initialize

To use the QueryDecomposition class, you first need to create an instance of the class.

from gritholdings.utils.query_decompose import QueryDecomposition

decomposer = QueryDecomposition()

Decompose

After the QueryDecomposition instance has been initialized, you can use the decompose method to decompose a query into multiple simpler questions. The decompose method takes a complex query as input and returns a list of simple queries.

query = "Is the customer's car still under warranty?"
decomposed_queries = decomposer.decompose(query)
print(decomposed_queries)
>>> ["When was the customer's vehicle purchased?", "What is the warranty period for the customer's vehicles?"]

If the input query is a simple question and doesn't require decomposition, the decompose method returns an empty list.

query = "What is your customer service phone number?"
decomposed_queries = decomposer.decompose(query)
print(len(decomposed_queries))
>>> 0