This is a sentence-transformers model finetuned from BAAI/bge-m3. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: BAAI/bge-m3
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8192, 'do_lower_case': False, 'architecture': '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()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("stel42/bs-faq-finetune-baai-bgem3")
# Run inference
queries = [
    "RSUSW itu singkatan dari apa?",
]
documents = [
    'RSU Syubbanul Wathon adalah kepanjangan dari RSUSW',
    'Lokasinya di Jl. Magelang - Kopeng km 08 Tegalrejo Kabupaten Magelang 56192',
    'Nama lain Siloam Hospitals Bangka Belitung adalah SHBB atau Siloam Pangkalan Baru',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.7324, 0.1836, 0.2039]])

Evaluation

Metrics

Triplet

Dataset: faq-test
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9768

Training Details

Training Dataset

Unnamed Dataset

Size: 2,718 training samples
Columns: query, answer_positive, and answer_negative

Approximate statistics based on the first 1000 samples:

	query	answer_positive	answer_negative
type	string	string	string
details	min: 5 tokens mean: 10.88 tokens max: 20 tokens	min: 8 tokens mean: 35.04 tokens max: 401 tokens	min: 8 tokens mean: 28.66 tokens max: 150 tokens

Samples:

query	answer_positive	answer_negative
`RS Jantung Diagram Cinere punya sebutan lain?`	`Nama lain RS Jantung Diagram Cinere adalah SHCN atau Siloam Cinere`	`Lokasinya di Jl. Kompol Maksum No.296, Peterongan, Kec. Semarang Sel., Kota Semarang, Jawa Tengah 50242`
`aplikasi mysiloam itu fungsinya apa`	`Aplikasi MySiloam adalah aplikasi mobile untuk membantu pengguna menikmati kemudahan layanan kesehatan yang disediakan oleh Rumah Sakit Siloam. Beberapa fitur yang sudah membantu banyak orang adalah, pembuatan janji temu dokter, pemesanan tes dan layanan kesehatan, emergency, dan ambulance call.`	`Lokasinya di JL. Mampang Prapatan XVI, Kel. Duren Tiga, Kec. Pancoran, Kota Adm. Jakarta Selatan, Prov. DKI Jakarta`
`Mau ke Siloam Surabaya, alamatnya apa?`	`Lokasinya di Jl. Raya Gubeng No.70, Gubeng, Kec. Gubeng, Kota SBY, Jawa Timur 60281`	`Ya, Anda dapat menanyakan kepada staff registrasi kami untuk mendapatkan perkiraan biaya sebelum perawatan Anda. Harap dicatat bahwa ada kemungkinan bahwa perkiraan biaya yang diberikan berbeda dari biaya akhir.`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false 5}

Evaluation Dataset

Unnamed Dataset

Size: 302 evaluation samples
Columns: query, answer_positive, and answer_negative
Approximate statistics based on the first 302 samples:
query answer_positive answer_negative
type string string string
details
min: 5 tokens
mean: 10.95 tokens
max: 18 tokens
min: 8 tokens
mean: 33.9 tokens
max: 255 tokens
min: 14 tokens
mean: 26.95 tokens
max: 54 tokens

Samples:

query	answer_positive	answer_negative
`SHPL ada di jalan apa?`	`Lokasinya di Jl. POM IX, Lorok Pakjo, Kec. Ilir Bar. I, Kota Palembang, Sumatera Selatan 30137`	`Pendaftaran MCU / Medical Check Up melalui Pembayaran pada pendaftaran MCU / Medical Check Up dapat dilakukan dengan transfer ke rekening Bank, Credit Card, dan pembayaran cash pada saat melakukan registrasi.`
`Dimana Siloam TB Simatupang?`	`Lokasinya di Jl. RA. Kartini No. 8, Cilandak Jakarta Selatan`	`Lama waktu yang dibutuhkan untuk menjalankan MCU / Medical Check Up sekitar 2-6 jam tergantung dari jenis paket pemeriksaan yang dijalankan.`
`siloam putera bahagia disebut apa lagi?`	`Nama lain Siloam Hospitals Putera Bahagia adalah SHCB atau Siloam Harjamukti`	`Lokasinya di Jl. Kelapa Dua Raya No.1001, Klp. Dua, Kec. Klp. Dua, Kabupaten Tangerang, Banten 15810`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false 5}

Training Logs

Epoch	Step	faq-test_cosine_accuracy
-1	-1	0.9768

Framework Versions

Python: 3.12.12
Sentence Transformers: 5.1.2
Transformers: 4.57.1
PyTorch: 2.8.0+cu126
Accelerate: 1.11.0
Datasets: 4.0.0
Tokenizers: 0.22.1

Citation

BibTeX

Sentence Transformers

bibtex

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MultipleNegativesRankingLoss

bibtex

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}