Sentence Transformers
The standard library for text/image embeddings, rerankers, and semantic search.
About
Sentence Transformers is the reference framework for computing embeddings: dense sentence embeddings, sparse encoders, ColBERT-style multi-vector (late interaction) encoders, and cross-encoder rerankers — all behind one SentenceTransformer API. It unlocks semantic search, similarity, paraphrase mining, and RAG retrieval, with 15,000+ pretrained models on the Hugging Face Hub (many topping the MTEB leaderboard) and first-class training/fine-tuning for custom data. Now officially maintained under the Hugging Face GitHub org (moved from UKPLab). Apache-2.0, 22.3M PyPI downloads/month.
What it can do
- **Dense embeddings**: 15,000+ pretrained models, MTEB-leaderboard SOTA options
- **Rerankers**: cross-encoder models for second-stage retrieval scoring
- **Sparse & multi-vector**: sparse encoders and ColBERT-style late-interaction retrieval
- **Multimodal**: image/text embedding extras (`[image]`)
- **Training**: fine-tune embedding, reranker, sparse, and multi-vector models on your data
- **Quantized inference**: ONNX/OpenVINO extras for CPU-speed embedding serving
Setup
```bash
pip install -U sentence-transformers
```
```python
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
embeddings = model.encode(["The cat sat on the mat.", "Dogs are great pets."])
print(embeddings.shape)
```
Requires PyTorch 2.2+ and transformers v5.0+ per current docs. Full docs: sbert.net.
Prerequisites
Python 3.10+; PyTorch 2.2+. `hf auth login` for higher Hub rate limits / gated models.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: 19,166 GitHub stars (log-scaled vs catalog peers); blended with ~22,332,616 monthly package downloads; last push 0d ago; 16 releases/12mo; 1332 open issues. Manual review: standard embeddings library; 15k+ Hub models, MTEB SOTA options; now under huggingface org (was UKPLab); retrieval quality depends on model choice, not the library. License: Apache-2.0.
★ 19,166 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.6/10
- Maintenance · recency, releases, issue load
- 8.2/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · 19,166 GitHub stars (log-scaled vs catalog peers)
- · blended with ~22,332,616 monthly package downloads
- · last push 0d ago; 16 releases/12mo; 1332 open issues
- · manual review: standard embeddings library; 15k+ Hub models, MTEB SOTA options; now under huggingface org (was UKPLab); retrieval quality depends on model choice, not the library
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.