ms-marco-MiniLM-L6-v2
| Property | Value |
|---|---|
| Author | cross-encoder |
| Model Type | Cross-Encoder |
| Performance (NDCG@10) | 74.30 |
| Speed | 1800 docs/sec on V100 GPU |
| Model Hub | Hugging Face |
What is ms-marco-MiniLM-L6-v2?
ms-marco-MiniLM-L6-v2 is a specialized cross-encoder model designed for passage ranking tasks, particularly optimized for the MS Marco dataset. It represents version 2 of the architecture, offering significant improvements over its predecessors in both performance and efficiency.
Implementation Details
The model implements a cross-encoder architecture with 6 layers (L6) of MiniLM, striking an optimal balance between computational efficiency and performance. It can be easily integrated using either SentenceTransformers or the Transformers library, supporting both high-level and low-level implementations.
- Achieves 74.30 NDCG@10 on TREC DL 19
- Processes 1800 documents per second on V100 GPU
- Delivers 39.01 MRR@10 on MS Marco Dev set
Core Capabilities
- Passage ranking and reranking
- Query-passage relevance scoring
- Information retrieval enhancement
- Compatible with major deep learning frameworks
Frequently Asked Questions
Q: What makes this model unique?
This model offers an excellent trade-off between performance and speed, outperforming many larger models while maintaining reasonable computational requirements. Its 6-layer architecture proves optimal for most passage ranking tasks.
Q: What are the recommended use cases?
The model excels in information retrieval scenarios where ranking or reranking of passages is required. It's particularly effective when used in conjunction with initial retrieval systems like ElasticSearch for re-ranking top results.
