HuggingFace - BAAI - BGE Reranker v2 m3

This repo provides a hands-on demonstration of using BAAl's (Beijing Academy of Artificial Intelligence) BGE-Reranker-v2-m3 model from Hugging Face for sequence reranking tasks. It includes two Jupyter notebooks:

• BGE_Reranker_Local.ipynb: For running the model locally using PyTorch and Hugging Face Transformers.
• BGE_Reranker_AzureML.ipynb: For deploying and running the model in Azure Machine Learning inference environment.

Note

Model weights are stored on the HuggingFace site and can be accessed at this model card page.

Table of Contents

• Prerequisites
• Part 1: Configuring Environment
• Part 2: BGE Reranker - Local use
• Part 3: BGE Reranker - Azure AI use
• Acknowledgements

Prerequisites

• Python 3.11+;
• GPU recommended for faster inference.

Part 1: Configuring Environment

1. Install necessary Python packages:

pip install -r requirements.txt

Part 2: BGE Reranker - Local use

This section covers the using of the BGE Reranker v2 m3 model locally with the BGE_Reranker_Local.ipynb notebook.

Key steps include:

1. Loading the Model: Import necessary libraries and load the BAAI/bge-reranker-v2-m3 model and tokenizer.
2. Creating a Reranking Function: Define a Python function rerank that takes a query and returns documents sorted by relevance score.
3. Testing: Demonstrate the rerank function with simple and more complex examples to show how the model's capability.

Sample reranking query results:

Query: What are the benefits of regular exercise?

Ranked Documents (most to least relevant):

1. Score: 6.7963
   Document: Regular exercise improves cardiovascular health, boosts mood, and helps maintain a healthy weight.

2. Score: 1.7936
   Document: Exercise has been shown to reduce the risk of chronic diseases such as diabetes and heart disease.

3. Score: 0.7743
   Document: Physical activity strengthens muscles and bones, and can improve sleep quality.

4. Score: -9.6014
   Document: Coffee contains caffeine which can improve alertness and concentration.

5. Score: -11.0410
   Document: The capital of France is Paris, which is known for its beautiful architecture.

Part 3: BGE Reranker - Azure AI use

With BGE_Reranker_AzureML.ipynb you can deploy and run the model on Azure Machine Learning online endpoint.

Key steps include:

1. Connecting to Azure ML: Authenticating and establishing a connection to your Azure Machine Learning workspace.
2. Finding the BGE Reranker Model: Identifying the baai-bge-reranker-v2-m3 model within the Hugging Face registry in Azure ML.
3. Deploying to Online Endpoint: Creating a managed online endpoint and deploying the BGE Reranker model to it.
4. Testing the Deployed Model: Sending sample query and text data to the deployed endpoint and processing the reranker's response.
5. Cleaning up Resources (Optional): Providing instructions for deleting the deployed endpoint and associated resources to avoid unnecessary costs.

Sample reranking query results:

================================================================================
PROCESSED RERANKING RESULTS:
================================================================================
Query: What is Deep Learning?
================================================================================
Rank Score      Text
================================================================================
1    0.999879   Deep learning is a subset of machine learning that uses neural network...
2    0.013742   Machine learning is a method of data analysis that automates analytica...
3    0.010945   Artificial intelligence is the simulation of human intelligence proces...
4    0.000142   Python is a high-level programming language widely used for web develo...

================================================================================

Acknowledgements

• BAAI for developing the BGE Reranker v2 m3 model;
• HuggingFace for hosting the model and providing the Transformers library;
• Microsoft Azure for cloud AI inference services.