sayqin/azerbaijani-sentence-embedding-v1

Azerbaijani Sentence Embedding Model (v1)

The first sentence embedding model for the Azerbaijani language, designed for semantic search and RAG pipelines.

Fine-tuned from allmalab/bert-base-aze using 50,000 sentence pairs mined from Azerbaijani Wikipedia (DOLLMA dataset) with MultipleNegativesRankingLoss.

⚠️ This is an early release (v1). Formal evaluation and benchmarking in progress.

Usage

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("YOUR_USERNAME/azerbaijani-sentence-embedding-v1")
embeddings = model.encode(["Bakı Azərbaycanın paytaxtıdır"])

Training

• Base model: allmalab/bert-base-aze
• Training data: 50,000 sentence pairs from Azerbaijani Wikipedia (DOLLMA)
• Loss function: MultipleNegativesRankingLoss
• Epochs: 3
• Batch size: 32

Citation

@inproceedings{isbarov-etal-2024-open,
    title = "Open foundation models for {A}zerbaijani language",
    author = "Isbarov, Jafar and Huseynova, Kavsar and Mammadov, Elvin and Hajili, Mammad and Ataman, Duygu",
    booktitle = "Proceedings of the First Workshop on Natural Language Processing for Turkic Languages (SIGTURK 2024)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand and Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.sigturk-1.2",
    pages = "18--28"

@misc{sayqin-2026-azerbaijani-embeddings,
  author       = {Sayqin Rustamli},
  title        = {Azerbaijani Sentence Embedding Model v1},
  address =    = {Strasbourg, France}
  year         = {2026},
  publisher    = {HuggingFace},
  howpublished = {\url{https://huggingface.co/sayqin/azerbaijani-sentence-embedding-v1}},
  note         = {First sentence embedding model for the Azerbaijani language}
}
}