Get the latest tech news
Introduction to Sentence Embeddings
Everything you wanted to know about sentence embeddings (and maybe a bit more)
For example, when someone wants to fine-tune a BERT model to perform text classification, a common approach is to add a linear layer on top of the[CLS] embedding. As we’ll see later in the blog post, the magnitude of the embeddings is not relevant when computing the cosine similarity, but it’s still a good think to normalize them in case we want to experiment with other ways to measure distances. Some of the players in this are are Microsoft (E5 models), Cohere, BAAI (BGE), Alibaba (GTE), NLP Group of The University of Hong Kong (Instructor), and Jina, among many others.
Or read this on Hacker News