J

Japanese Reranker Tiny V2

Developed by hotchpotch
This is a very compact and fast Japanese reranking model, suitable for improving the accuracy of RAG systems and can run efficiently on CPUs or edge devices.
Downloads 339
Release Time : 5/7/2025

Model Overview

This model is a Japanese text reranker, primarily used to reorder retrieved documents to improve relevance. Based on the ModernBert architecture, it is specially optimized for performance in resource-constrained environments.

Model Features

Lightweight and Efficient
Only a 3-layer architecture, capable of running at practical speeds on CPUs or Apple Silicon environments.
Resource-Friendly
Improves the accuracy of RAG systems without requiring expensive GPU resources.
Edge Device Compatible
Suitable for deployment on edge devices or production environments with high latency requirements.
Optimized Inference
Supports Flash Attention 2 acceleration and ONNX quantization optimization.

Model Capabilities

Japanese text relevance scoring
Retrieval result reranking
Fast inference

Use Cases

Information Retrieval
Document Retrieval Optimization
Reranks search engine results to improve relevance
Achieved a score of 0.6455 on the JQaRA dataset
Question Answering Systems
QA Candidate Answer Sorting
Reranks candidate answers generated by a QA system for relevance
Achieved a score of 0.9608 on the JSQuAD dataset
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase