gte-reranker-modernbert-base
| Property | Value |
|---|---|
| Model Size | 149M parameters |
| Max Sequence Length | 8192 tokens |
| Primary Language | English |
| Author | Alibaba-NLP (Tongyi Lab) |
| Model Type | Text Reranker |
| BEIR Score | 56.19 |
| LoCo Score | 90.68 |
What is gte-reranker-modernbert-base?
The gte-reranker-modernbert-base is an advanced text reranking model developed by Alibaba's Tongyi Lab. Built on the modernBERT architecture, it's specifically designed for improving search and retrieval systems by reranking candidate documents based on their relevance to a query. The model demonstrates exceptional performance across various benchmarks, particularly in long document retrieval (LoCo) and BEIR evaluation tasks.
Implementation Details
The model can be easily implemented using either the transformers or sentence-transformers libraries. It supports efficient Flash Attention 2 for improved performance and can handle sequences up to 8192 tokens in length. The model outputs relevance scores for query-document pairs, making it ideal for search result refinement and document ranking tasks.
- Built on modernBERT foundation model architecture
- Supports both transformers and sentence-transformers implementations
- Compatible with Flash Attention 2 for enhanced efficiency
- Outputs calibrated relevance scores for text pairs
Core Capabilities
- Long document processing (up to 8192 tokens)
- State-of-the-art performance on LoCo benchmark (90.68)
- Strong BEIR evaluation scores (56.19)
- Efficient cross-encoder architecture for precise relevance scoring
- Robust performance across various retrieval tasks
Frequently Asked Questions
Q: What makes this model unique?
The model's key strengths lie in its ability to handle long documents effectively and its superior performance on reranking tasks. Built on the modernBERT architecture, it achieves impressive scores on LoCo (90.68) and BEIR (56.19) benchmarks while maintaining a relatively compact size of 149M parameters.
Q: What are the recommended use cases?
The model is ideal for improving search systems, document retrieval applications, and any scenario requiring precise ranking of text pairs. It's particularly effective for long document retrieval, question-answering systems, and information retrieval tasks where accurate relevance scoring is crucial.
