| Property | Value |
|---|---|
| Parameter Count | 435M |
| Model Type | Cross-Encoder Reranker |
| License | Apache 2.0 |
| Tensor Type | FP16 |
What is mxbai-rerank-large-v1?
mxbai-rerank-large-v1 is the flagship model in Mixedbread AI's reranker family, designed to significantly improve search relevance. Built on a DeBERTa-v2 architecture, this 435M parameter model achieves state-of-the-art performance with 48.8 NDCG@10 on BEIR benchmarks, outperforming both traditional lexical search and competitive embedding models like cohere-embed-v3.
Implementation Details
The model implements a cross-encoder architecture optimized for reranking tasks. It can be easily integrated using sentence-transformers or accessed through Mixedbread's API. The model processes query-document pairs simultaneously to produce highly accurate relevance scores, making it ideal for refining search results.
- Optimized for FP16 inference
- Supports both Python and JavaScript implementations
- Includes built-in truncation and padding handling
- Compatible with sentence-transformers ecosystem
Core Capabilities
- Achieves 74.9% Accuracy@3 on benchmark datasets
- Excels at reranking both keyword and semantic search results
- Supports batch processing for efficient inference
- Provides flexible API integration options
Frequently Asked Questions
Q: What makes this model unique?
The model combines superior accuracy with practical deployment capabilities, outperforming larger models while maintaining reasonable computational requirements. Its performance on BEIR benchmarks (48.8 NDCG@10) sets it apart from existing solutions.
Q: What are the recommended use cases?
The model excels at improving search quality in various scenarios: enhancing keyword search results, refining semantic search outputs, and providing more relevant document rankings in information retrieval systems. It's particularly effective when combined with existing search infrastructure.