Updating the embedding model for semantic indexing

BMC HelixGPT supported only intfloat/multilingual-e5-base model for semantic indexing. From version 25.3.00,  the sentence-transformers/paraphrase-multilingual-mpnet-base-v2 model is also supported. For enhanced performance, we recommend using the new model. This model improves document chunking and semantic indexing, leading to more accurate and relevant responses from BMC HelixGPT.

Important

  • By default, the intfloat/multilingual-e5-base is the active model.
  • Only one model can be active at a time.
  • Switching models requires re-indexing the existing content.

Before you begin

Make sure to create a clone of the existing index.

Task 1: To clone the existing index

  1. Log in to BMC Helix Innovation Studio.
  2. In the Workspace tab, go to HelixGPT Agent Studio.
  3. From the Records, select the DataConnectionJob and click Edit data.
  4. In the Data editor (DataConnectionJob) window, click the job you want to clone.
  5. In the Edit record window, specify the following fields:
    Field nameDescription
    TypeSelect Clone Index
    SourceSemanticIndexIdSpecify the ID of the currently active model
    TargetSemanticIndexIdSpecify the ID of the model you want to activate
    ExecuteNow.Enable the radio button

    Sematic_index_connection.png

  6. Click Save.

Task 2: To activate the embedding model for semantic indexing

  1. Log in to BMC Helix Innovation Studio.
  2. On the Workspace tab, click HelixGPT Agent Studio.
  3. From the Records, select the SemanticIndex and click Edit data.
  4. In the Data editor (SemanticIndex) window, click the model you want to activate.
  5. In the Edit record window, from the Status drop-down list, select Active.
    Sematic_index_connection1.png
  6. Click Save.
    The new model is activated, and the status of a previously active model automatically changes to Enabled.

 

Tip: For faster searching, add an asterisk to the end of your partial query. Example: cert*