README.md

---
tags:
- sentence-transformers
- sentence-similarity
- feature-extraction
- telepix
pipeline_tag: sentence-similarity
library_name: sentence-transformers
---
<p align="center">
    <img src="https://cdn-uploads.huggingface.co/production/uploads/61d6f4a4d49065ee28a1ee7e/V8n2En7BlMNHoi1YXVv8Q.png" width="400"/>
<p>

# PIXIE-Rune-v1.0
**PIXIE-Rune-v1.0** is an encoder-based embedding model trained on Korean and English triplets, developed by [TelePIX Co., Ltd](https://telepix.net/). 
**PIXIE** stands for Tele**PIX** **I**ntelligent **E**mbedding, representing TelePIX’s high-performance embedding technology. 
The model is multilingual, specifically optimized for both Korean and English. 
It demonstrates strong performance on retrieval tasks in both languages, achieving robust results across a wide range of Korean- and English-language benchmarks. 
This makes it well-suited for real-world applications that require high-quality semantic search in Korean, English, or both.

## Model Description
- **Model Type:** Sentence Transformer
<!-- - **Base model:** [Unknown](https://huggingface.co/unknown) -->
- **Maximum Sequence Length:** 8192 tokens
- **Output Dimensionality:** 1024 dimensions
- **Similarity Function:** Cosine Similarity
<!-- - **Training Dataset:** Unknown -->
- **Language:** Bilingual — optimized for high performance in Korean and English
- **License:** apache-2.0 

### Full Model Architecture

```
SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)
```

## Quality Benchmarks
**PIXIE-Rune-v1.0** is a multilingual embedding model specialized for Korean and English retrieval tasks. 
It delivers consistently strong performance across a diverse set of domain-specific and open-domain benchmarks in both languages, demonstrating its effectiveness in real-world semantic search applications.
The table below presents the retrieval performance of several embedding models evaluated on a variety of Korean and English benchmarks.
We report **Normalized Discounted Cumulative Gain (NDCG)** scores, which measure how well a ranked list of documents aligns with ground truth relevance. Higher values indicate better retrieval quality.  
- **Avg. NDCG**: Average of NDCG@1, @3, @5, and @10 across all benchmark datasets.  
- **NDCG@k**: Relevance quality of the top-*k* retrieved results.
 
#### Korean Retrieval Benchmarks
Our model, **telepix/PIXIE-Rune-v1.0**, achieves state-of-the-art performance across most metrics and benchmarks, demonstrating strong generalization across domains such as multi-hop QA, long-document retrieval, public health, and e-commerce.

| Model Name | # params | Avg. NDCG | NDCG@1 | NDCG@3 | NDCG@5 | NDCG@10 |
|------|:---:|:---:|:---:|:---:|:---:|:---:|
| **telepix/PIXIE-Rune-v1.0** | 568M | **0.6905** | **0.6461** | **0.6859** | **0.7063** | **0.7238** |
|  |  |  |  |  |  |  |
| nlpai-lab/KURE-v1 | 568M | 0.6751 | 0.6277 | 0.6725 | 0.6907 | 0.7095 |
| dragonekue/BGE-m3-ko | 568M | 0.6658 | 0.6225 | 0.6627 | 0.6795 | 0.6985 |
| Snowflake/snowflake-arctic-embed-l-v2.0 | 568M | 0.6592 | 0.6118 | 0.6542 | 0.6759 | 0.6949 |
| BAAI/bge-m3 | 568M | 0.6573 | 0.6099 | 0.6533 | 0.6732 | 0.6930 |
| Qwen/Qwen3-Embedding-0.6B | 595M | 0.6321 | 0.5894 | 0.6274 | 0.6455 | 0.6662 |
| jinaai/jina-embeddings-v3 | 572M | 0.6293 | 0.5800 | 0.6254 | 0.6456 | 0.6665 |
| Alibaba-NLP/gte-multilingual-base | 305M | 0.6111 | 0.5542 | 0.6089 | 0.6302 | 0.6511 |
| openai/text-embedding-3-large | N/A | 0.6015 | 0.5466 | 0.5999 | 0.6187 | 0.6409 |

Descriptions of the benchmark datasets used for evaluation are as follows:
- **Ko-StrategyQA**  
  A Korean multi-hop open-domain question answering dataset designed for complex reasoning over multiple documents.
- **AutoRAGRetrieval**  
  A domain-diverse retrieval dataset covering finance, government, healthcare, legal, and e-commerce sectors.
- **MIRACLRetrieval**  
  A document retrieval benchmark built on Korean Wikipedia articles.
- **PublicHealthQA**  
  A retrieval dataset focused on medical and public health topics.
- **BelebeleRetrieval**  
  A dataset for retrieving relevant content from web and news articles in Korean.
- **MultiLongDocRetrieval**  
  A long-document retrieval benchmark based on Korean Wikipedia and mC4 corpus.
- **XPQARetrieval**  
  A real-world dataset constructed from user queries and relevant product documents in a Korean e-commerce platform.

#### English Retrieval Benchmarks
Our model, **telepix/PIXIE-Rune-v1.0**, achieves strong performance on a wide range of tasks, including fact verification, multi-hop question answering, financial QA, and scientific document retrieval, demonstrating competitive generalization across diverse domains.
 
| Model Name | # params | Avg. NDCG | NDCG@1 | NDCG@3 | NDCG@5 | NDCG@10 |
|------|:---:|:---:|:---:|:---:|:---:|:---:|
| **telepix/PIXIE-Rune-v1.0** | 568M | **0.5781** | **0.5691** | **0.5663** | **0.5791** | **0.5979** |
|  |  |  |  |  |  |  |
| Snowflake/snowflake-arctic-embed-l-v2.0 | 568M | 0.5812 | 0.5725 | 0.5705 | 0.5811 | 0.6006 |
| Qwen/Qwen3-Embedding-0.6B | 595M | 0.5558 | 0.5321 | 0.5451 | 0.5620 | 0.5839 |
| Alibaba-NLP/gte-multilingual-base | 305M | 0.5541 | 0.5446 | 0.5426 | 0.5574 | 0.5746 |
| BAAI/bge-m3 | 568M | 0.5318 | 0.5078 | 0.5231 | 0.5389 | 0.5573 |
| dragonekue/BGE-m3-ko | 568M | 0.5307 | 0.5125 | 0.5174 | 0.5362 | 0.5566 |
| nlpai-lab/KURE-v1 | 568M | 0.5272 | 0.5017 | 0.5171 | 0.5353 | 0.5548 |

Descriptions of the benchmark datasets used for evaluation are as follows:
- **ArguAna**  
  A dataset for argument retrieval based on claim-counterclaim pairs from online debate forums.
- **FEVER**  
  A fact verification dataset using Wikipedia for evidence-based claim validation.
- **FiQA-2018**  
  A retrieval benchmark tailored to the finance domain with real-world questions and answers.
- **HotpotQA**  
  A multi-hop open-domain QA dataset requiring reasoning across multiple documents.
- **MSMARCO**  
  A large-scale benchmark using real Bing search queries and corresponding web documents.
- **NQ**  
  A Google QA dataset where user questions are answered using Wikipedia articles.
- **SCIDOCS**  
  A citation-based document retrieval dataset focused on scientific papers.
  
## Usage

### Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

```bash
pip install -U sentence-transformers
```

Then you can load this model and run inference.
```python
from sentence_transformers import SentenceTransformer
  
# Load the model
model_name = 'PIXIE-Rune-v1.0'
model = SentenceTransformer(model_name)

# Define the queries and documents
queries = [
    "텔레픽스는 어떤 산업 분야에서 위성 데이터를 활용하나요?",
    "국방 분야에 어떤 위성 서비스가 제공되나요?",
    "텔레픽스의 기술 수준은 어느 정도인가요?",
]
documents = [
    "텔레픽스는 국방, 농업, 자원, 해양 등 다양한 분야에서 위성 데이터를 분석하여 서비스를 제공합니다.",
    "정찰 및 감시 목적의 위성 영상을 통해 국방 관련 정밀 분석 서비스를 제공합니다.",
    "TelePIX의 광학 탑재체 및 AI 분석 기술은 Global standard를 상회하는 수준으로 평가받고 있습니다.",
    "텔레픽스는 우주에서 수집한 정보를 분석하여 '우주 경제(Space Economy)'라는 새로운 가치를 창출하고 있습니다.",
    "텔레픽스는 위성 영상 획득부터 분석, 서비스 제공까지 전 주기를 아우르는 솔루션을 제공합니다.",
]

# Compute embeddings: use `prompt_name="query"` to encode queries!
query_embeddings = model.encode(queries, prompt_name="query")
document_embeddings = model.encode(documents)

# Compute cosine similarity scores
scores = model.similarity(query_embeddings, document_embeddings)

# Output the results
for query, query_scores in zip(queries, scores):
    doc_score_pairs = list(zip(documents, query_scores))
    doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True)
    print("Query:", query)
    for document, score in doc_score_pairs:
        print(score, document)

```

### Framework Versions
- Python: 3.10.16
- Sentence Transformers: 4.0.1
- Transformers: 4.51.3
- PyTorch: 2.6.0+cu124
- Accelerate: 1.5.2
- Datasets: 2.21.0
- Tokenizers: 0.21.1


## Contact

If you have any suggestions or questions about this Model, please reach out to the authors at bmkim@telepix.net.