Epicure consists of three sibling skip-gram ingredient embeddings retrained on a multilingual recipe corpus. The dataset includes 4.14 million recipes sourced from 11 different platforms and covers seven languages.