Sentence-Transformers

Sentence-Transformers, also known as SBERT, is a Python framework for computing text and image embeddings and reranking scores with neural models. It covers three model families: dense embedding models for semantic similarity, cross-encoders that score text pairs for reranking search results, and sparse encoders for efficient retrieval. More than 15,000 compatible pretrained models are available on the Hugging Face Hub spanning over 100 languages, many ranking well on the MTEB embeddings benchmark, and the library provides training APIs with more than 20 loss functions for fine-tuning embedding models and around 10 for rerankers. Typical applications include semantic search, similarity scoring, paraphrase mining, clustering, duplicate detection, and retrieval-augmented generation pipelines. The project began at the Ubiquitous Knowledge Processing Lab at TU Darmstadt and is now maintained at Hugging Face; it installs with pip, requires Python 3.10 or newer with PyTorch, and is open source under the Apache 2.0 license.