KoVRE: Korean Visual Document Retrieval Embedding

This repository provides the training and evaluation code for KoVRE, a compact 2B single-vector embedding model for Korean visual document retrieval. KoVRE is trained in two stages: contrastive learning on Korean and English query–document pairs, followed by reranker-based knowledge distillation on Korean data.

Installation

uv venv --python 3.10
uv sync

Quick start

Before running the scripts, update the configuration variables for your environment.

Stage 1: Contrastive learning

Train KoVRE on Korean and English visual document retrieval data with mined hard negatives:

bash scripts/train.sh

Stage 2: Knowledge distillation

Continue training the Stage 1 checkpoint on Korean data using reranker scores as soft targets. Set MODEL_NAME_OR_PATH in scripts/distill.sh to the Stage 1 checkpoint before running:

bash scripts/distill.sh

Evaluation

Evaluate a checkpoint on KoViDoRe and SDS KoPub VDR at multiple embedding dimensions. Set MODEL_NAME_OR_PATH in scripts/evaluate.sh to the checkpoint to evaluate:

bash scripts/evaluate.sh

Performance

Results are reported in nDCG@10. AVG is the average across the four KoViDoRe domains, and OVR additionally includes SDS KoPub VDR.

Model	Params	KoViDoRe					SDSKoPub	OVR
		Cybersecurity	Economic	Energy	HR	AVG
jina-clip-v2	0.9B	0.1993	0.0011	0.1096	0.0294	0.0849	0.0732	0.0825
jina-v5-omni-nano	1B	0.4404	0.0640	0.2017	0.0695	0.1939	0.0961	0.1743
jina-v5-omni-small	2B	0.4357	0.0744	0.2380	0.1029	0.2128	0.1902	0.2082
Qwen3-VL-Embedding-2B	2B	0.6111	0.1592	0.4123	0.1842	0.3417	0.4285	0.3591
Qwen3-VL-Embedding-8B	8B	0.7809	0.2373	0.6360	0.3613	0.5039	0.7293	0.5489
jina-embeddings-v4 (single-vector)	4B	0.7280	0.2058	0.6273	0.4106	0.4929	0.7222	0.5388
jina-embeddings-v4 (multi-vector)	4B	0.7714	0.2359	0.6752	0.4799	0.5406	0.7802	0.5885
KoVRE (Stage 1)	2B	0.7444	0.2797	0.6506	0.5002	0.5437	0.7214	0.5792
KoVRE (Stage 1 + Stage 2)	2B	0.7627	0.2987	0.6576	0.5082	0.5568	0.7324	0.5919

License

Apache-2.0

Contact

For questions or suggestions, please open an issue on the GitHub repository or contact the maintainer:

• Yongbin Choi - whybe.choi@gmail.com

Citation

If you use KoVRE in your research, please cite this work.

@article{choi2026kovre,
  title={KoVRE: Training an Efficient Embedding Model for Korean Visual Document Retrieval},
  author={Choi, Yongbin and Shim, Gyuho and Jang, Youngjoon},
  journal={arXiv preprint arXiv:2608.01389},
  year={2026}
}