LlamaIndex re-ranker and Elasticsearch re-ranker: Comparison review

This notebook is based on the article LlamaIndex re-ranker and Elasticsearch re-ranker: Comparison review and has the main goal to compare the performance of the LlamaIndex re-ranker and the Elasticsearch re-ranker.

Dataset

Setup keys

Elasticsearch client

Mappings

The mappings are defined to create the index in Elasticsearch. The metadata field is a nested field that contains the product information. The semantic_field is a field that is used to store the semantic representation of the product. The embeddings field is a dense vector field that is used to store the embeddings of the product used to perform the re-ranking with LlamaIndex.

User question

LlamaIndex Reranking

In this section we will use the LlamaIndex re-ranker to re-rank the results of the search query.

Indexing data to Elasticsearch

The documents are indexed to Elasticsearch using the ElasticsearchStore class. This is a wrapper around the Elasticsearch client and inserts the documents into the Elasticsearch index in a metadata field.

LLM setup

LlamaIndex rerank feature

Without rerank

With rerank

Elasticsearch Reranking

In this section we will use the Elasticsearch re-ranker to re-rank the results of the search query.

Creating inference rerank endpoint

Function to execute search queries

Semantic query

Cleaning environment

Delete the resources used to prevent them from consuming resources.