A framework that fine-tunes transformer models to produce meaningful embeddings of entire sentences or paragraphs, rather than just individual tokens.