Skip to main content

Uploading and Training Indices

Starting the Upload Process​

  1. Click the Action icon of an index with a Not Uploaded status and choose Upload from the context menu.

    Upload a new index

  2. Select the train type for your index. Click NEXT and follow your train type specific uploading instructions.

    Select Train Type

Flat Uploading​

  1. Enter the parameters below. Parameters vary based on training type.

    Flat Parameters

    ParameterDetails
    Train TypeOptimize the dataset during the train process using Optuna or Grid. Note that this will take longer to train.
    Dataset N BitsThe number of features for which the vector will be in its binary format after quantization.

  1. In the Prefilter window, choose which filters you would like to use when searching through your dataset. Only fields of type keyword are available as filters.

    Prefilter

  2. Verify your choices and click SUBMIT.

    Flat Summary

IVF Uploading​

  1. Enter the parameters below. Parameters vary based on training type.

    IVF Parameters

    ParameterDetails
    Train TypeOptimize the dataset during the train process using Optuna or Grid. Note that this will take longer to train.
    Dataset N BitsThe number of features for which the vector will be in its binary format after quantization.
    Cluster N BitsThe number of features for which the centroids vectors will be in their binary format after quantization
    Number of ClustersThe number of clusters in the IFV structure. If you select Boards, then the number of clusters will be automatically determined based on the maximum possible.

  1. In the Prefilter window, choose which filters you would like to use when searching through your dataset. Only fields of type keyword are available as filters.

    Prefilter

  2. Verify your choices and click SUBMIT.

    IVF Summary

IVF-HNSW Uploading​

  1. Enter the parameters below. Parameters vary based on training type.

    IVF-HNSW Parameters

    ParameterDetails
    Use SSDUse SSD for memory efficiency.
    M - number of edgesEquivalent to m. The number of neighbors for each node in the HNSW graph.
    Number of ClustersEquivalent to ef_construction. The number of clusters in the IFV-HNSW structure. If you select Boards, then the number of clusters will be automatically determined based on the maximum possible.

  1. In the Prefilter window, choose which filters you would like to use when searching through your dataset. Only fields of type keyword are available as filters.

    Prefilter

  2. Verify your choices and click SUBMIT.

    IVF-HNSW Summary