DPR Question Encoder (Single NQ Base)
| Property | Value |
|---|---|
| License | CC-BY-NC-4.0 |
| Paper | Dense Passage Retrieval for Open-Domain Question Answering |
| Base Architecture | BERT-base-uncased |
| Training Data | Natural Questions Dataset |
What is dpr-question_encoder-single-nq-base?
This is a specialized BERT-based encoder model designed for open-domain question answering as part of Facebook's Dense Passage Retrieval (DPR) framework. It specifically handles the question encoding component, transforming natural language questions into dense vector representations that can be efficiently matched with relevant passages.
Implementation Details
The model utilizes a BERT-base architecture fine-tuned on the Natural Questions dataset. It works in conjunction with a passage encoder to create a robust retrieval system, achieving impressive accuracy rates of 78.4% for top-20 and 85.4% for top-100 passage retrieval.
- Trained using 8 32GB GPUs for optimal performance
- Implements dense encoding for efficient similarity matching
- Integrates with FAISS for fast nearest neighbor search
Core Capabilities
- Question encoding into dense vector representations
- Optimized for open-domain question answering
- Efficient retrieval across large document collections
- High accuracy in passage retrieval tasks
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in converting questions into dense vector representations, enabling efficient similarity-based retrieval from large document collections. It's specifically optimized for the Natural Questions dataset and achieves state-of-the-art performance in open-domain QA tasks.
Q: What are the recommended use cases?
The model is best suited for building open-domain question answering systems, information retrieval applications, and search engines where efficient question-to-passage matching is required. It should be used alongside a corresponding passage encoder for complete functionality.
