Configure fused vectors

更新时间:
复制 MD 格式

Fused vectors combine multiple fields, such as text and images, into a single vector by using a multimodal model. This enables cross-modal retrieval capabilities such as text-to-image and image-to-text searches in e-commerce and multimodal search scenarios.

Usage notes

The fused vector feature requires engine version 1.5.8 or later. If your current engine version is lower, contact technical support to request an upgrade.

Supported models

The fused vector feature calls a multimodal model service to vectorize field content. Each model has specific limits on the number and type of input fields:

Model source

Model name

Output dimensions

Input limits

Model Studio

qwen3-vl-embedding

Supported dimensions:

  • 256

  • 512

  • 768

  • 1024

  • 1536

  • 2048

  • 2560

Each request supports up to 20 content inputs, including a maximum of 5 images.

AI Search Open Platform

ops-gme-qwen2-vl-2b-instruct

1536

Each request supports one image and one text input.

ops-mm-embedding-v1-2b

1536

Each request supports one image and one text input.

ops-mm-embedding-v1-7b

3584

Each request supports one image and one text input.

ops-mm-embedding-ecom-001

128

Each request supports one image and one text input.

Configure fused vectors

Fused vectors are configured in the Add table Field Configuration step.

  1. Navigate to the Field Configuration page and add the fields for fusion to the field list. A fused vector requires at least two participating fields.

    If a participant field contains image content, you must first configure a data source for it. In the Data Source column of the field list, select the data type (text or image) and configure the source based on your data storage method. Supported data sources for the image type include URL, Base64 encoding, Object Storage Service (OSS), and DLF-Object Table.

  2. Click Data Processing at the bottom of the page. From the Data Processing Template drop-down list, select Fused Vector.

  3. In the Participant Fields section, select at least two fields to include in the fused vector encoding.

  4. In the Services section, configure the model service:

    • Model: Select a multimodal model from the drop-down list. The default model is Model Studio qwen3-vl-embedding.

    • Field Name: The system automatically generates an output field name for the fused vector, such as _fused_vector. You do not need to modify this name.

  5. In the API Key field, enter the API Key for the selected model service.

  6. Read the risk and billing notices, and then select the checkboxes.

  7. Click OK to complete the fused vector configuration.

After the configuration is complete, the system automatically adds the fused vector field to the field list with the following characteristics:

  • The Data Source column displays the names of the source fields used for the fused vector encoding.

  • The Field Type column shows a Fused identifier, indicating the field was generated by the fused vector process.

  • In the Actions column, click Edit to reopen the data processing configuration panel and display the previously saved settings.

Other data processing templates

In addition to Fused Vector, the data processing feature supports the following templates:

Template name

Description

Dense Vectorization

Converts a text field into a dense vector.

Dense + Sparse Vectorization

Generates both a dense vector and a sparse vector to enable hybrid search.

Image Vectorization

Converts an image field into a vector.

Image Content Analysis

Parses text and other content from an image.

Image Content Analysis + Image Vectorization

Parses image content and then vectorizes it.

Video Processing

Processes video field data.

Field data source configuration

On the Data Source page, Data Source is a separate column.

  • Default fields from scenario templates: These fields have a preset data type that you cannot modify.

  • User-added fields: You can select either text or image as the data type.

    For the Image type, supported data sources are: URL, Base64 encoding, Object Storage Service (OSS), and DLF-Object Table.