| Property | Value |
|---|---|
| Parameter Count | 109M |
| Base Model | intfloat/e5-base-v2 |
| License | CC BY-NC-SA 4.0 |
| Paper | arXiv:2410.05077 |
What is zebra-retriever-e5-base-v2?
zebra-retriever-e5-base-v2 is a specialized retrieval model designed for zero-shot commonsense question answering. It serves as a crucial component in the ZEBRA framework, which enhances LLM performance through example-based retrieval augmentation. The model is built upon the e5-base-v2 architecture and optimized for retrieving relevant question-knowledge pairs from large collections.
Implementation Details
The model implements a three-stage pipeline approach: example retrieval, knowledge generation, and informed reasoning. It operates in F32 tensor format and is designed to work seamlessly with the ZEBRA framework for enhanced question answering capabilities.
- Retrieval-based architecture optimized for question-answer pair matching
- Integration with knowledge generation systems
- Support for zero-shot learning scenarios
- Compatibility with various LLM backends
Core Capabilities
- Example-based retrieval for commonsense questions
- Integration with language models for knowledge generation
- Support for multiple choice question answering
- Performance improvements across various QA benchmarks
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its specialized ability to retrieve relevant examples for commonsense reasoning, showing significant performance improvements across multiple benchmarks like CSQA, ARC-C, and PIQA. It's particularly effective when combined with larger language models in the ZEBRA framework.
Q: What are the recommended use cases?
The model is ideal for enhancing commonsense question answering systems, particularly in scenarios requiring zero-shot capabilities. It's specifically designed for retrieval augmentation in educational, general knowledge, and commonsense reasoning applications.