BAAI/bge-reranker-large¶
Model Information¶
The BAAI/bge-reranker-large is a cross-encoder reranking model developed by the Beijing Academy of Artificial Intelligence (BAAI). It is designed to re-rank top-k documents retrieved by initial retrieval models, enhancing the relevance of search results. This model is particularly effective in applications such as search engines, question answering, and information retrieval systems.
- Model Developer: Beijing Academy of Artificial Intelligence (BAAI)
- Model Release Date: March 18, 2024
- Supported Languages: English, Chinese
Model Architecture¶
- Base Model: XLM-RoBERTa-large
- Architecture Type: Transformer-based cross-encoder
- Input Format: Concatenated query and document pairs
- Output: Relevance score indicating the similarity between the query and document
Benchmark Scores¶
BAAI/bge-reranker-large delivers strong reranking performance across common retrieval benchmarks.
| Dataset | Metric | Score | Note |
|---|---|---|---|
| MS MARCO | MRR@10 | 40.2 | Dev set |
| TREC DL '19 | NDCG@10 | 71.6 | Document reranking |
| BEIR (avg) | NDCG@10 | 59.3 | Avg. across 18 datasets |
| LoTTE (EN) | MRR@10 | 52.1 | Open-domain QA reranking |
Evaluated using FlagEmbedding pipeline with Hugging Face Transformers.