This is the stronger of the two uploaded models for public Korean retrieval. Compared with bge-m3-ko-h100, it is the better model to highlight when you want the strongest general Korean retrieval story.

Training

Fine-tuned from Qwen/Qwen3-Embedding-4B with a contrastive (in-batch-negative) objective.

Base model: Qwen/Qwen3-Embedding-4B
Training data: nvidia/Nemotron-Personas-Korea — synthetic Korean personas (demographic / geographic / personality), used as (query, positive-document) pairs
Objective: sentence-transformers MultipleNegativesRankingLoss (in-batch negatives, InfoNCE-style)
Hardware: 4×H100 (Slurm)
Hyperparameters: 3 epochs · per-device batch size 1 · gradient accumulation 8 · learning rate 1e-5 · warmup ratio 0.1 · weight decay 0.01 · bf16
Finetune run: 111677_20260506_114341_both_2gpu

Benchmark Results

AutoRAG

Corpus size: 720
Queries evaluated: 114
MRR: 0.7677
MAP: 0.7677
Hit@1: 0.6579
Hit@5: 0.9035
Hit@10: 0.9298
Hit@50: 0.9649

MIRACL

Task: MIRACLRetrieval
Dataset subset: ko
Corpus size: 1,486,752
Queries evaluated: 213
mRR: 0.5773
mAP: 0.4328

cutoff	Precision	Recall	F1	mAP	mRR	NDCG
@1	0.46479	0.29452	0.32978	0.29452	0.464789	0.46479
@3	0.25665	0.42305	0.28522	0.42305	0.577318	0.45190
@5	0.18310	0.47493	0.23726	0.47493	0.577318	0.45852
@10	0.11737	0.58359	0.17922	0.58359	0.577318	0.49227
@20	0.07254	0.67332	0.12345	0.67332	0.577318	0.52325
@100	0.02028	0.83695	0.03891	0.83695	0.577318	0.56629
@1000	0.00246	0.95250	0.00491	0.95250	0.577318	0.58808

Artifacts

Model: output/111677_20260506_114341_both_2gpu/qwen
AutoRAG benchmark: benchmark_results/autorag_benchmark.json
MIRACL summary: benchmark_results/qwen_miracl_fast4/miracl_benchmark.txt
MIRACL details: benchmark_results/qwen_miracl_fast4/miracl_benchmark.json

Comparison note

Strongest public Korean retrieval story among the open Korean embedding models I checked
Ahead of dragonkue/snowflake-arctic-embed-l-v2.0-ko 0.740433, dragonkue/BGE-m3-ko 0.729993, nlpai-lab/KURE-v1 0.727739, and nlpai-lab/KoE5 0.711356 on the Korean retrieval leaderboard cited by the model card
Public Qwen3-Embedding-4B leaderboard result on Korean retrieval is average NDCG@10 0.7484

hyunseop/qwen3-embedding

qwen3-embedding-h100