ms-marco-MiniLM-L-12-v2
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Downloads | 643,654 |
| Processing Speed | 960 docs/sec on V100 GPU |
| Performance | NDCG@10: 74.31, MRR@10: 39.02 |
What is ms-marco-MiniLM-L-12-v2?
ms-marco-MiniLM-L-12-v2 is a powerful cross-encoder model specifically designed for passage ranking tasks. It represents the latest version (v2) in the MS Marco series, trained on the Microsoft Marco Passage Ranking dataset. This model achieves state-of-the-art performance while maintaining reasonable processing speeds.
Implementation Details
The model utilizes the MiniLM architecture with 12 layers, offering an optimal balance between performance and efficiency. It can be easily implemented using either the Transformers library or SentenceTransformers framework, with built-in support for query-passage pair scoring.
- Supports both PyTorch and JAX frameworks
- Optimized for sequence classification tasks
- Maximum sequence length of 512 tokens
- Processes approximately 960 documents per second on V100 GPU
Core Capabilities
- High-accuracy passage ranking with NDCG@10 score of 74.31
- Efficient query-passage pair scoring
- Seamless integration with popular NLP frameworks
- Support for batch processing of multiple query-passage pairs
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its exceptional balance between performance and speed, achieving top-tier NDCG@10 scores while maintaining reasonable processing speeds. It's particularly noteworthy for being part of the v2 series, which shows significant improvements over the original versions.
Q: What are the recommended use cases?
The model excels in information retrieval tasks, particularly for re-ranking passages after initial retrieval. It's ideal for applications requiring precise query-document matching, such as search engines, question-answering systems, and document retrieval systems.
