mirror of
https://github.com/StanGirard/quivr.git
synced 2024-12-02 08:40:53 +03:00
0559075cd3
This pull request adds a new guide to the documentation that explains how to configure reranking in Quivr. The guide covers the steps to configure the reranking service using the FlashrankRerank library and the Cohere ReRanker. It also provides information on the required environment variables and the default behavior of the Ranker.
29 lines
1.0 KiB
Plaintext
29 lines
1.0 KiB
Plaintext
---
|
|
title: Reranking
|
|
description: Learn how to configure reranking in Quivr.
|
|
---
|
|
|
|
Quivr uses its own reranking service to improve the quality of search results. This guide will show you how to configure the reranking service in Quivr.
|
|
|
|
## Reranking Configuration
|
|
|
|
We are using FlashrankRerank from [Flashrank](https://github.com/PrithivirajDamodaran/FlashRank) to improve the quality of search results.
|
|
It is a reranking open-source library that uses deep learning models to rerank search results.
|
|
|
|
## Use Cohere ReRanker
|
|
|
|
To use the Cohere ReRanker, you need to configure the following environment variables in your `.env` file:
|
|
|
|
```bash
|
|
COHERE_API_KEY=your-cohere-api-key
|
|
```
|
|
|
|
By default, the `COHERE_API_KEY` variable is not set. If you want to use the Cohere ReRanker, you need to set it to your Cohere API key.
|
|
|
|
## Configuration
|
|
|
|
By default the Ranker outputs the best 5 results.
|
|
|
|
We only use the Cohere ReRanker to improve the quality of search results. We do not collect any personal data or sensitive information.
|
|
|