ms-marco-TinyBERT-L-2
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Downloads | 77,642 |
| Processing Speed | 9,000 docs/sec |
| NDCG@10 (TREC DL 19) | 67.43 |
| MRR@10 (MS Marco Dev) | 30.15 |
What is ms-marco-TinyBERT-L-2?
ms-marco-TinyBERT-L-2 is a lightweight cross-encoder model specifically designed for passage ranking tasks. Trained on the MS Marco Passage Ranking dataset, it represents a careful balance between efficiency and performance, offering impressive processing speeds while maintaining competitive ranking accuracy.
Implementation Details
The model implements a cross-encoder architecture optimized for query-passage pair scoring. It can be easily integrated using either the Transformers library or SentenceTransformers framework, with the latter offering a more streamlined implementation process.
- Achieves 67.43 NDCG@10 on TREC Deep Learning 2019
- Processes 9,000 documents per second on V100 GPU
- Implements efficient TinyBERT architecture with 2 layers
- Supports both PyTorch and JAX frameworks
Core Capabilities
- Fast and efficient passage ranking
- Query-passage pair scoring
- Batch processing support
- Integration with popular NLP frameworks
- Support for padding and truncation
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its exceptional processing speed (9,000 docs/sec) while maintaining reasonable performance metrics, making it ideal for large-scale information retrieval systems where speed is crucial.
Q: What are the recommended use cases?
The model is best suited for information retrieval tasks, particularly in scenarios requiring fast passage re-ranking. It's ideal for applications where you need to sort passages based on relevance to a query, especially in production environments with high throughput requirements.
