---
library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- matryoshka
- multilingual
- embeddings
language:
- multilingual
- en
- ar
- de
- es
- fr
- zh
- ru
- tr
- ko
- ja
license: cc-by-nc-4.0
base_model: xlm-roberta-base
---

# Matryoshka Text Embedding v1

**Matryoshka Embedding Model with Flexible Dimensions**

This is a [sentence-transformers](https://www.SBERT.net) model with Matryoshka Representation Learning,
allowing flexible dimension truncation from 64D to 1024D.

## Model Details

- **Model Type**: Sentence Transformer with Matryoshka Representation Learning
- **Base Architecture**: XLM-RoBERTa
- **Dimensions**: 64, 128, 256, 384, 512, 768, 1024
- **Max Sequence Length**: 8192 tokens
- **Languages**: 100+ languages
- **Output Dimensionality**: 1024 (with 7 truncation options)

## Usage
```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer('matryoshka-text-embedding-v1')

# Full precision
embeddings = model.encode(["Hello World"])

# Optimized for production (recommended)
embeddings = model.encode(["Hello World"], truncate_dim=512)
```

See [README.md](README.md) for detailed documentation.

## Performance (Self-Reported)

| Task | Metric | Score |
|------|--------|-------|
| SciFact | NDCG@10 | 0.6308 |
| STS17 | Spearman | 0.8096 |
| STSBenchmark | Spearman | 0.8506 |

## License

CC-BY-NC-4.0 - See LICENSE file for details and acknowledgments.

## Acknowledgments

This model builds upon:
- **XLM-RoBERTa**: Base architecture
- **BAAI**: RetroMAE and BGE-M3 research contributions
- **Matryoshka Representation Learning**: Training methodology

## Citation

```bibtex
@misc{matryoshka-text-embedding-v1,
  title={Matryoshka Text Embedding v1},
  author={Hasan Kurşun and Kerem Berkay Yanık},
  year={2025},
  url={https://huggingface.co/matryoshka-text-embedding-v1},
  organization={Lumees},
  contact={hello@lumees.io},
  website={https://lumees.io}
}
```