Glossary

Text Embedding

A text embedding is a numerical vector representation of a sentence or passage that captures semantic meaning, produced by an embedding model and used to power similarity search in retrieval-augmented generation.

A text embedding converts a sentence or passage into a vector of real numbers positioned so that passages with similar meaning sit close together in that vector space, which is what lets a system retrieve semantically related content instead of only exact keyword matches. watsonx.ai exposes embeddings through its /ml/v1/text/embeddings REST endpoint and a Python Embeddings class, accepting up to 1,000 lines of text per request. IBM's own encoder models include granite-embedding-278m-multilingual (768 dimensions, 512 max input tokens) and the older slate-125m-english-rtrvr-v2 and slate-30m-english-rtrvr-v2, both scheduled for withdrawal on August 8, 2026; Microsoft's multilingual-e5-large is also supported. For an IBM i shop, watsonx.ai is the source of these vectors, but as of September 2026 Db2 for i has no native vector store to hold them, so embeddings typically land in OpenSearch running on Power instead.

Related Terms