ms-marco-MiniLM-L-2-v2
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Architecture | Cross-Encoder MiniLM |
| Performance (NDCG@10) | 71.01 |
| Processing Speed | 4100 docs/sec |
What is ms-marco-MiniLM-L-2-v2?
ms-marco-MiniLM-L-2-v2 is a specialized cross-encoder model trained on the MS Marco Passage Ranking dataset. It represents the second version of a compact yet efficient architecture designed for information retrieval tasks, particularly excelling in passage re-ranking scenarios.
Implementation Details
The model utilizes the MiniLM architecture with 2 layers, offering an optimal balance between performance and speed. It can be easily implemented using either the Transformers library or SentenceTransformers framework, supporting batch processing and providing normalized relevance scores for query-passage pairs.
- Achieves 71.01 NDCG@10 on TREC DL 19
- Processes 4100 documents per second on V100 GPU
- Supports maximum sequence length of 512 tokens
Core Capabilities
- Efficient passage re-ranking for search systems
- Query-passage relevance scoring
- Integration with existing search pipelines
- Batch processing support
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its exceptional balance between performance and speed, offering significant improvements over the first version while maintaining efficient processing capabilities. It's particularly well-suited for production environments where both accuracy and speed are crucial.
Q: What are the recommended use cases?
The model is ideal for search result re-ranking, document retrieval systems, and any application requiring efficient assessment of query-document relevance. It's particularly effective when used in conjunction with initial retrieval systems like ElasticSearch for re-ranking top results.
