ms-marco-MiniLM-L12-v2
| Property | Value |
|---|---|
| Author | cross-encoder |
| Model Type | Cross-Encoder |
| Performance (NDCG@10) | 74.31 |
| Processing Speed | 960 docs/sec |
| Model Hub | Hugging Face |
What is ms-marco-MiniLM-L12-v2?
ms-marco-MiniLM-L12-v2 is a sophisticated cross-encoder model specifically designed for information retrieval tasks, particularly excelling in passage ranking. As version 2 of the architecture, it represents a significant improvement over its predecessors, achieving state-of-the-art performance metrics while maintaining reasonable computational efficiency.
Implementation Details
The model implements a cross-encoder architecture optimized for the MS Marco Passage Ranking task. It can be easily integrated using either SentenceTransformers or the Transformers library, offering flexibility in implementation. The model processes query-passage pairs to generate relevance scores, enabling effective document ranking.
- Achieves 74.31 NDCG@10 on TREC DL 19
- MRR@10 score of 39.02 on MS Marco Dev
- Processing speed of 960 documents per second on V100 GPU
- 12-layer architecture optimized for efficiency
Core Capabilities
- Passage ranking and reranking
- Query-document relevance scoring
- Information retrieval optimization
- Integration with search systems
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its optimal balance between performance and efficiency, achieving top-tier NDCG@10 scores while maintaining reasonable processing speeds. It's particularly noteworthy for being a v2 improvement over previous versions, with enhanced accuracy metrics.
Q: What are the recommended use cases?
The model is ideal for implementing retrieve & re-rank systems, particularly when working with large document collections. It's especially effective when combined with initial retrieval systems like ElasticSearch, where it can be used to re-rank initially retrieved passages for improved relevance.
