ms-marco-electra-base
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Downloads | 447,746 |
| Processing Speed | 340 docs/sec on V100 GPU |
| Task Type | Text Classification, Passage Ranking |
What is ms-marco-electra-base?
ms-marco-electra-base is a specialized cross-encoder model designed for passage ranking tasks, built on the ELECTRA architecture. It was specifically trained on the MS Marco Passage Ranking dataset, offering robust performance for information retrieval applications with a balanced trade-off between accuracy and processing speed.
Implementation Details
The model implements a cross-encoding architecture optimized for query-passage pair scoring. It can be easily integrated using either the Transformers library or SentenceTransformers framework, supporting batch processing and providing normalized relevance scores for passage ranking.
- Achieves 71.99 NDCG@10 on TREC DL 19 evaluation
- Delivers 36.41 MRR@10 on MS Marco Dev set
- Processes approximately 340 documents per second on V100 GPU
Core Capabilities
- Query-passage relevance scoring
- Passage re-ranking for search results
- Information retrieval optimization
- Support for both PyTorch and Transformers frameworks
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely combines ELECTRA's efficient architecture with MS Marco training, offering a sweet spot between computational efficiency and ranking accuracy. It provides better performance than BERT-base models while maintaining reasonable inference speed.
Q: What are the recommended use cases?
The model is ideal for search engine result re-ranking, document retrieval systems, and any application requiring precise relevance scoring between queries and text passages. It's particularly effective when used in a two-stage retrieval setup where initial candidates are first retrieved using faster methods.
