DPR Context Encoder (Single-NQ-Base)
| Property | Value |
|---|---|
| Developer | |
| License | CC-BY-NC-4.0 |
| Base Architecture | BERT-base-uncased |
| Research Paper | Dense Passage Retrieval for Open-Domain Question Answering |
| Top-20 Accuracy (NQ) | 78.4% |
What is dpr-ctx_encoder-single-nq-base?
This model is a specialized context encoder component of Facebook's Dense Passage Retrieval (DPR) system, specifically trained on the Natural Questions dataset. It's designed to encode text passages into dense vector representations for efficient open-domain question answering.
Implementation Details
The model leverages BERT's base architecture to transform text passages into dense vector representations. It works in conjunction with a question encoder to enable efficient retrieval of relevant passages for given queries.
- Built on BERT-base-uncased architecture
- Trained specifically on Natural Questions dataset
- Optimized for passage encoding in retrieval tasks
- Uses FAISS for efficient similarity search during inference
Core Capabilities
- Dense passage encoding for retrieval tasks
- Achieves 85.4% accuracy in top-100 retrieval on NQ dataset
- Efficient representation learning for text passages
- Seamless integration with DPR question encoder
Frequently Asked Questions
Q: What makes this model unique?
This model specializes in encoding text passages into dense vectors, optimized specifically for open-domain question answering. Its training on the Natural Questions dataset makes it particularly effective for real-world search queries.
Q: What are the recommended use cases?
The model is best suited for building open-domain QA systems, particularly when combined with its companion question encoder. It's ideal for applications requiring efficient passage retrieval from large document collections.
