OptionaldimensionThe dimension of the embedding vectors. Defaults to the model's default dimension. Models that support several dimensions accept any of them; for example, llama-text-embed-v2 accepts 512 as well as its default of 1024. A dimension the model does not support is rejected, and the error lists the supported ones.
Refer to the model guide for each model's supported dimensions.
OptionalmetricThe distance metric used for similarity search. Defaults to the model's default metric. A metric the model does not support is rejected.
Possible values: cosine, dotproduct, or euclidean.
OptionalmodelThe name of a hosted dense embedding model. Defaults to llama-text-embed-v2. A sparse model is rejected; declare it under sparse_embed instead.
Refer to the model guide for available models.
OptionalreadModel-specific parameters applied when embedding search query text, such as truncate. Validated against the model; omitted parameters take the model's defaults. The resolved parameters are returned when the index is described.
OptionalwriteModel-specific parameters applied when embedding document text, such as truncate. Validated against the model; omitted parameters take the model's defaults. The resolved parameters are returned when the index is described.
Dense embedding configuration for a
stringfield. Pinecone embeds the field's text with a hosted embedding model when documents are upserted or updated, and embeds the query text with the same model when a search scores the field with anembedscoring method.Every key is optional;
{}selects the default model at its default dimension and metric. The server resolves each omitted key from the model, validates each supplied one against it, and returns the resolved configuration when the index is described. The configuration is fixed at creation and cannot be changed afterwards.An
embedfield occupies the index's single dense vector slot, so it cannot be combined with adense_vectorfield or withembedon another field.Export
CreateEmbedConfig