SentenceTransformer

Description

Sentence Transformers (a.k.a. SBERT) is a Python module for accessing, using, and training state-of-the-art embedding and reranker models. It can be used to compute embeddings using Sentence Transformer models or to calculate similarity scores using Cross-Encoder (reranker) models. This unlocks a wide range of applications.

Features

• Model Capabilities:
  • Compute embeddings with Sentence Transformer models.
  • Calculate similarity scores with Cross-Encoder (reranker) models.
• Applications:
  • Semantic search
  • Semantic textual similarity
  • Paraphrase mining
• Model Availability & Customization:
  • Access to over 10,000 pre-trained Sentence Transformers models are available on Hugging Face.
  • Ability to train or finetune custom embedding models.
  • Ability to train or finetune custom reranker models.
• API & Utilities:
  • Includes a new training API for CrossEncoder (reranker) models (introduced in v4.0).
  • Supports ONNX and OpenVINO backends for CrossEncoder models to speed up inference (introduced in v4.1).
  • Provides functions for:
    • Quantizing embeddings: quantize_embeddings()
    • Semantic search: semantic_search(), semantic_search_faiss(), semantic_search_usearch()
    • Community detection: community_detection()
    • Mining hard negatives: mine_hard_negatives()
    • Normalizing embeddings: normalize_embeddings()
    • Paraphrase mining: paraphrase_mining()
    • Truncating embeddings: truncate_embeddings()
    • Exporting models: export_dynamic_quantized_onnx_model(), export_optimized_onnx_model(), export_static_quantized_openvino_model()
    • Similarity metrics: cos_sim(), dot_score(), euclidean_sim(), manhattan_sim(), pairwise_cos_sim(), pairwise_dot_score(), pairwise_euclidean_sim(), pairwise_manhattan_sim()

Installation

• Install using pip.
• Requires Python 3.9+ and PyTorch 1.11.0+.