KoReranker¶
Korean-specific document reranking.
Overview¶
| Field | Value |
|---|---|
| Type | CrossEncoder (Korean) |
| Library | torch, transformers |
| Default Model | Dongjin-kr/ko-reranker |
Installation¶
pip install "autorag-research[gpu]"
# or
uv add "autorag-research[gpu]"
Configuration¶
_target_: autorag_research.rerankers.koreranker.KoRerankerReranker
model_name: Dongjin-kr/ko-reranker
Options¶
| Option | Type | Default | Description |
|---|---|---|---|
| model_name | str | Dongjin-kr/ko-reranker |
KoReranker model name |
| max_length | int | 512 | Maximum input sequence length |
| device | str | None | Device (auto-detected) |
| batch_size | int | 64 | Batch size for multiple queries |
Usage¶
from autorag_research.rerankers import KoRerankerReranker
reranker = KoRerankerReranker()
results = reranker.rerank("RAG란 무엇인가?", ["문서1", "문서2", "문서3"], top_k=2)
When to Use¶
Specifically designed for Korean language documents. Use this when:
- Your corpus is primarily Korean text
- You need Korean-aware relevance scoring