Member Junction
    Preparing search index...

    Interface MJAIModelEntity_EmbeddingConfigurationSettings

    Embedding knobs, read at call time by the embedding resolvers (AIEngine.EmbedText, AIEmbeddingRunner). Set them on the catalog layers (MJ: AI Models, MJ: AI Model Vendors). Asymmetric models such as EmbeddingGemma 2 or Gemini Embedding 2 are trained to read a different instruction prefix on a search query than on a stored document; the resolvers prepend the one that matches the call's inputType, so drivers and callers never hardcode model-specific strings. Providers with a native input-type parameter (Cohere) leave these unset.

    interface MJAIModelEntity_EmbeddingConfigurationSettings {
        DocumentPrefix?: string;
        QueryPrefix?: string;
    }
    Index

    Properties

    DocumentPrefix?: string

    Prepended to a text embedded as a stored document (inputType: 'document', the default), e.g. 'title: none | text: '.

    QueryPrefix?: string

    Prepended to a text embedded as a search query (inputType: 'query'), e.g. 'task: search result | query: '.