Polish Reranker Bge V2
This is a reranking model based on BAAI/bge-reranker-v2-m3 and further fine-tuned on a large-scale Polish text pair dataset, supporting long-context processing.
Downloads 549
Release Time : 9/25/2024
Model Overview
This model is a Polish text reranking model based on the BAAI/bge-reranker-v2-m3 architecture, fine-tuned on Polish datasets, particularly suitable for ranking tasks involving long documents.
Model Features
Long context support
Supports processing texts up to 8192 tokens, making it suitable for ranking tasks involving long documents.
Efficient inference
Utilizes Flash Attention 2 technology, achieving up to 400% speed improvement in processing long texts.
Polish language optimization
Specifically fine-tuned for Polish, delivering excellent performance in Polish information retrieval tasks.
Knowledge distillation
Uses BAAI/bge-reranker-v2.5-gemma2-lightweight as the teacher model for knowledge distillation.
Model Capabilities
Text relevance scoring
Information retrieval result reranking
Long document processing
Use Cases
Information retrieval
Search engine result optimization
Reranks search engine results to improve relevance
Achieved NDCG@10 of 64.21 on the Polish Information Retrieval Benchmark (PIRB)
Document retrieval system
Processes retrieval systems containing long documents to optimize ranking results
Particularly suitable for processing long documents up to 8192 tokens
Featured Recommended AI Models