https://github.com/leo310/rag-chunking-evaluation

Assess the effectiveness of chunking strategies in RAG systems via a custom evaluation framework.
https://github.com/leo310/rag-chunking-evaluation

chunking evaluation-framework retrieval retrieval-augmented-generation

Last synced: 3 months ago
JSON representation

Assess the effectiveness of chunking strategies in RAG systems via a custom evaluation framework.

• Host: GitHub
• URL: https://github.com/leo310/rag-chunking-evaluation
• Owner: Leo310
• License: mit
• Created: 2024-06-30T10:45:23.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2024-10-17T09:12:22.000Z (over 1 year ago)
• Last Synced: 2025-09-04T22:08:25.925Z (7 months ago)
• Topics: chunking, evaluation-framework, retrieval, retrieval-augmented-generation
• Language: Jupyter Notebook
• Homepage:
• Size: 4.44 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE
  • Citation: CITATION.cff

Awesome Lists containing this project

README

          
# RAG Chunking Evaluation

This repository contains code and datasets for evaluating chunking strategies in Retrieval-Augmented Generation (RAG) systems. The project includes various benchmarks, data loaders, and utility functions to facilitate the evaluation process.

## Setup

1. **Clone this repository**

2. **Create a virtual environment:**

   ```sh

   python -m venv venv

   source venv/bin/activate  # On Windows use `venv\Scripts\activate`

   ```

3. **Install dependencies:**

   ```sh

   pip install -r requirements.txt

   ```

4. **Set up environment variables:**

   Copy `.env.example` to `.env` and fill in the required values.

## Usage

Follow the instructions in the `my_benchmark` notebook to run the proposed chunking evaluation framework. The specific chunking strategies under evaluation are detailed in the `chunking_strategies` notebook.

Each step in the evaluation pipeline generates intermediate results, which are saved in the `data` directory for later review and loading.

The `experimental` directory includes tests for other benchmarks and evaluation frameworks, such as Ragas, Trulens, and Multi-Hop-RAG.