BAAI/bge-reranker-v2-gemma
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:2.6BQuant:BF16Ctx Length:8kLicense:apache-2.0Architecture:Transformer0.1K Open Weights Warm

BAAI/bge-reranker-v2-gemma is a 2.6 billion parameter multilingual reranker model developed by BAAI, built upon the Google Gemma-2B architecture. This model is designed to assess the relevance between a query and a passage, outputting a similarity score. It excels in multilingual contexts, demonstrating strong performance in both English proficiency and general multilingual capabilities for information retrieval tasks.

Loading preview...

Overview

BAAI/bge-reranker-v2-gemma is a 2.6 billion parameter multilingual reranker model, initialized from Google's Gemma-2B. Unlike traditional embedding models, this reranker takes a question and a document as input and directly outputs a relevance score, which can be mapped to a float value between 0 and 1 using a sigmoid function. It is part of the FlagEmbedding project, focusing on efficient and effective information retrieval.

Key Capabilities

  • Multilingual Reranking: Optimized for diverse language contexts, performing well across multiple languages, including strong English proficiency.
  • Relevance Scoring: Computes a direct relevance score between a query and a passage, indicating how well the passage answers the query.
  • LLM-based Architecture: Leverages the Gemma-2B large language model as its foundation, contributing to its robust performance.
  • Fine-tuning Support: Provides scripts and data format guidelines for fine-tuning the model on custom datasets, allowing for adaptation to specific use cases.

Good For

  • Improving Search Results: Enhancing the ranking of retrieved documents in information retrieval systems.
  • Multilingual Applications: Ideal for scenarios requiring relevance scoring across various languages.
  • Resource-constrained Environments: Offers a balance of performance and efficiency, suitable for deployment where computational resources are a consideration.
  • Research and Development: A strong baseline for further research into reranking techniques and multilingual NLP.