lightblue/lb-reranker-0.5B-v1.0

The lightblue/lb-reranker-0.5B-v1.0 is a 0.5 billion parameter reranker model, fine-tuned from a Qwen/Qwen2.5-0.5B-Instruct checkpoint by Lightblue. It is designed to determine the relatedness of a query to a text, outputting a score between 1 and 7. This model excels in multilingual retrieval tasks, having been trained on data in over 95 languages, and demonstrates competitive performance against state-of-the-art rerankers while being optimized for efficient inference.