BGE Reranker Large
| Property | Value |
|---|---|
| Parameter Count | 560M |
| License | MIT |
| Languages | Chinese, English |
| Framework | PyTorch, ONNX |
What is bge-reranker-large?
BGE Reranker Large is a powerful cross-encoder model designed for high-accuracy text similarity scoring and reranking. Unlike traditional embedding models, it directly processes query-document pairs to produce relevance scores, making it ideal for refining search results and improving information retrieval systems.
Implementation Details
The model utilizes a cross-encoder architecture that performs full-attention over input pairs, offering superior accuracy compared to bi-encoder approaches. It's built on XLM-RoBERTa architecture and supports both PyTorch and ONNX inference frameworks.
- Optimized for both Chinese and English text processing
- Supports efficient top-k reranking of search results
- Achieves state-of-the-art performance on multiple benchmarks including MTEB and C-MTEB
- Includes FP16 support for faster inference
Core Capabilities
- Cross-lingual reranking with support for Chinese-English and English-Chinese pairs
- Direct relevance scoring without need for separate embeddings
- High performance on medical QA tasks (CMedQAv1/v2)
- Efficient integration with existing search pipelines
Frequently Asked Questions
Q: What makes this model unique?
The model combines high accuracy with practical efficiency, using a cross-encoder architecture that directly computes relevance scores for text pairs. It achieves superior performance on both monolingual and cross-lingual tasks, making it especially valuable for multilingual applications.
Q: What are the recommended use cases?
The model is best used for reranking top-k results from a first-stage retrieval system. It's particularly effective for improving search quality in medical QA systems, document retrieval, and cross-lingual information retrieval tasks.
