---
license: mit
---

This repo contains the code needed to run AIR-Bench using llamaindex.

In this repo, I implement a custom retriever that uses `text-embedding-ada-002` as the dense embedding model, bm25 for sparse embeddings, and the `QueryFusionRetriever()` to combine results from these two, as well as generating extra queries for retrieval.

## Usage

`pip install llama-index llama-index-retrievers-bm25`

`python ./run_airbench.py`

## Customization

Feel free to use this as a template to evaluate other llama-index retrieval pipelines! You just need to customize the setup in `create_retriever_fn()` with the setup that you want to test against.