bge-m3-ko-h100
H100 fine-tuned Korean embedding model based on dragonkue/BGE-m3-ko.
This is the better fit for the local AutoRAG corpus among the two uploaded models. Compared with qwen3-embedding-h100, it is the stronger model to highlight for the domain-specific benchmark.
Training
Fine-tuned from dragonkue/BGE-m3-ko (the official BAAI/bge-m3 continued-trained on Korean MIRACL) with a contrastive (in-batch-negative) objective.
- Base model:
dragonkue/BGE-m3-ko— Korean-specialized BGE-M3 - Training data:
nvidia/Nemotron-Personas-Korea— synthetic Korean personas (demographic / geographic / personality), used as(query, positive-document)pairs - Objective:
sentence-transformersMultipleNegativesRankingLoss(in-batch negatives, InfoNCE-style) - Hardware: 4×H100 (Slurm)
- Hyperparameters: 3 epochs · per-device batch size 8 · gradient accumulation 4 · learning rate 2e-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: 0.7773
- MAP: 0.7773
- Hit@1: 0.6754
- Hit@5: 0.9123
- Hit@10: 0.9474
- Hit@50: 0.9825
MIRACL
- Task:
MIRACLRetrieval - Main score: 0.50
- Dataset revision:
9c09abc13478308c27598f350e31d8f06b9b5481
| cutoff | Precision | Recall | F1 | mAP | mRR | NDCG |
|---|---|---|---|---|---|---|
| @1 | 0.39906 | 0.25093 | 0.30812 | 0.25093 | 0.399061 | 0.39906 |
| @3 | 0.26604 | 0.43788 | 0.33099 | 0.34631 | 0.50313 | 0.43568 |
| @5 | 0.20282 | 0.52623 | 0.29279 | 0.37510 | 0.523787 | 0.45984 |
| @10 | 0.13380 | 0.64972 | 0.22190 | 0.40375 | 0.537201 | 0.50233 |
| @20 | 0.08052 | 0.72989 | 0.14504 | 0.41724 | 0.540598 | 0.53096 |
| @100 | 0.02272 | 0.90287 | 0.04432 | 0.43030 | 0.543259 | 0.57932 |
| @1000 | 0.00254 | 0.98881 | 0.00507 | 0.43149 | 0.543447 | 0.59457 |
Artifacts
- Model:
output/111677_20260506_114341_both_2gpu/bge - AutoRAG benchmark:
benchmark_results/autorag_benchmark.json - MIRACL summary:
benchmark_results/miracl_benchmark.txt - MIRACL details:
benchmark_results/mteb/MIRACLRetrieval.json
Comparison note
- Strongest AutoRAG result of the two uploaded models
- Weaker MIRACL-KO result than
rtfin-qwen3-embedding-h100 - Below
dragonkue/snowflake-arctic-embed-l-v2.0-koandrtfin-qwen3-embedding-h100on the public Korean retrieval leaderboard, but still a strong local-domain model - Good to cite when the audience cares about the local domain corpus rather than the widest public Korean retrieval leaderboard