bge-reranker-v2-m3 Now Available on Novita AI

Reranking is a cornerstone of modern search systems, empowering applications to deliver highly relevant results through sophisticated relevance analysis. This guide explores what rerankers (reranking models) are, why they are crucial for search quality, and how BAAI/bge-reranker-v2-m3 perform on elevating search capabilities to drive superior user experiences.

Understanding Rerankers

A reranker is a specialized AI model designed to refine and reorder search results by evaluating the semantic relevance between queries and candidate documents. Unlike traditional retrieval systems that focus on speed and recall, rerankers prioritize precision by computing detailed relevance scores for query-document pairs.

In modern AI applications, rerankers play a particularly crucial role in RAG (Retrieval-Augmented Generation) systems, where the quality of retrieved documents directly impacts the final generated response. The reranking stage acts as a precision filter, ensuring that only the most relevant documents proceed to the content generation phase.

Modern information retrieval systems typically employ a two-stage architecture as shown above. The first stage uses initial fast retrieval methods like dual-tower models, BM25, or vector databases to quickly filter candidate documents from large collections. The second stage employs rerankers to perform fine-grained sorting of these candidate results, focusing on precise relevance judgment.

Rerankers use a cross-encoder architecture that processes query and document jointly, enabling deeper interaction modeling compared to traditional retrieval methods. This architecture allows rerankers to identify complex semantic associations, synonyms, and hierarchical relationships that keyword-based matching might miss.

Applications of Rerankers

Rerankers excel in various application scenarios where semantic understanding and precise ranking are crucial:

Critical Domain Applications

• Medical information systems: Retrieving relevant medical literature where accuracy is life-critical
• Legal document retrieval: Finding precise case law and statutes for legal professionals
• Scientific research: Surfacing the most relevant studies from thousands of papers

AI-Powered Systems

• Question-Answering Systems: In RAG systems, rerankers ensure that the most semantically relevant documents are selected as context for answer generation, directly impacting response quality and accuracy
• Recommendation Systems: Sort candidate items by relevance to user preferences, improving engagement and personalization

Enterprise Solutions

• Enterprise Search: Large organizations use rerankers to optimize document retrieval across vast knowledge bases, making information discovery more efficient and accurate
• Search Engines: Perform secondary ranking of search results to improve relevance and user satisfaction beyond traditional keyword matching

Cross-Platform Applications

• Multilingual Platforms: Handle cross-language retrieval scenarios effectively, providing consistent search quality across different languages

Understanding BAAI/bge-reranker-v2-m3

BAAI/bge-reranker-v2-m3 is a lightweight multilingual reranker developed by the Beijing Academy of Artificial Intelligence (BAAI). As part of the BGE (Beijing General Embedding) series, this model is specifically optimized for reranking tasks across multiple languages.

Key Features:

• Lightweight design: 568M parameters for efficient deployment
• Multilingual support: 18+ languages with cross-language capabilities
• Fast inference: Millisecond-level response times
• Easy integration: Multiple APIs and libraries available

Technical Specifications:

• Architecture: Cross-encoder sequence classification
• Input format: [query text, document text]
• Maximum length: 512 tokens
• Output: Direct relevance scores with optional normalization
• Acceleration: fp16/bf16 support for GPU inference

Supported Languages: English, Chinese, Japanese, Korean, Spanish, French, German, Russian, Arabic, Hindi, Bengali, Persian, Finnish, Indonesian, Thai, Telugu, Swahili, Yoruba

The model is trained on diverse datasets including bge-m3-data (multilingual retrieval), Quora train data (question-answer pairs), and FEVER train data (fact verification), ensuring robust performance across various domains and use cases.

llama-index.

BEIR.

CMTEB-retrieval.

miracl (multi-language).

How to Access the BAAI/bge-reranker-v2-m3 on Novita AI

Novita AI is an AI cloud platform that offers developers an easy way to deploy AI models using our simple API, while also providing the affordable and reliable GPU cloud for building and scaling.

Step 1: Log In and Access the Model Console

Log in to your account and access the model console.

Step 2: Choose Your Model and Start a Free Trail

Browse through the available options and search for the model that suits your needs.

Step 3: Get Your API Key

To authenticate with the API, we will provide you with a new API key. Entering the “Settings“ page, you can copy the API key as indicated in the image.

Step 4: Install the API

Install API using the package manager specific to your programming language.

After installation, import the necessary libraries into your development environment. Initialize the API with your API key to start interacting with Novita AI rerank service. This example demonstrates how to use the rerank API.

curl --request POST \
  --url "https://api.novita.ai/v3/openai/rerank" \
  --header "Authorization: Bearer <Your API Key>" \
  --header "Content-Type: application/json" \
  --data @- << 'EOF'
{
  "model": "baai/bge-reranker-v2-m3",
  "query": "Who is Novita",
  "documents": [
    "Novita AI is an all-in-one AI cloud solution that empowers businesses with open-source model APIs, serverless GPUs, and on-demand GPU instances. Drive innovation and gain a competitive edge with the power of Novita AI.",
    "Access Novita AI's Model Library with 200+ APIs for language, image, audio, and video. Simplify AI deployment with powerful, scalable solutions.",
    "Optimize your AI with Novita's GPU instances and serverless GPU cloud. Save up to 50%, auto-scale, and access high-capacity storage for global deployment."
  ],
  "top_n": 3
}
EOF
 

Upon registration, Novita AI provides a $0.5 credit to get you started!

If the free credits is used up, you can pay to continue using it.

Rerankers play a critical role in information retrieval and AI applications, allowing systems to effectively understand semantic relevance and improve search accuracy. Advanced models like BAAI/bge-reranker-v2-m3 elevate these capabilities through superior performance and extensive language support, making them indispensable tools for driving modern AI applications.

Frequently Asked Questions

Novita AI is an AI cloud platform that offers developers an easy way to deploy AI models using our simple API, while also providing the affordable and reliable GPU cloud for building and scaling.