Retrieval testing

Retrieval testing is used to identify similar chunks in response to a query, determining which data chunks will be processed by the language model. Users can submit queries related to the data stored in the knowledge base. The system then retrieves relevant chunks of information, ranking them based on how well they match the user's query. The most relevant chunk, with the highest matching score, is displayed first, followed by chunks with lower scores.

By analyzing the returned chunks, users can assess how well the LLM interprets and retrieves relevant information for various queries.

1. To perform retrieval testing, navigate to the ‘Retrieval Testing’ tab. Enter a query related to your data in the provided text box, and click the ‘Test’ button to execute the test.

1. You will be provided with a retrieval result, which is an array of chunks from the knowledge base

1. You can find a minimum matching score for chunks to filter search results based on relevance.

2. In the History section, you can view logs of previous queries and interactions. These logs include details such as the source, text, date, and time of each interaction.