The DeepSeek Large Model is an intelligent data processing tool within the AI-Large Model Processing Operator. Through custom model configuration, generation strategies, and persona settings, it enables unstructured data parsing, multilingual translation, and other functions, making it suitable for enterprise intelligent data processing scenarios. This document introduces how to use the DeepSeek Large Model operator.

• You have activated the DeepSeek Large Model service or other Custom large model services.
• The enterprise administrator has configured the large model for the operator. For details, please refer to Configure Large Model Invocation.

Step 1: Create a DeepSeek Large Model Operator #

1. Click 'Data Integration', select 'Visual Knowledge Processing', click 'Create', and choose 'Create Offline Task'.
   
2. Complete Task Creation according to the steps in Offline Task Creation. When configuring processing nodes, select the DeepSeek Large Model operator from 'AI-Large Model Processing' to add to the Data Connection.
   

Step 2: Set Model Integration Parameters #

Click 'Custom Model', and in the 'Model Settings' popup, configure the relevant parameters.


The relevant parameter descriptions are as follows:

Step 3: Configure Model Persona Settings and Response Logic #

Users can customize the large model's persona settings, functions, and workflows.

Use natural language to define the large model's persona settings, functions, and workflows. You may refer to the following content for custom configuration:

# Role
You are a precise and efficient intelligent delivery address resolution assistant robot that can accurately and efficiently parse input address information into structured content. For any input address, the parsing result must be quickly and accurately returned in JSON format.{
  /** Province */
  "province": "string",
  /** City */
  "city": "string",
  /** District */
  "region": "string",
  /** Detailed Address */
  "address": "string",
  /** Recipient */
  "people": "string",
  /** Phone */
  "tel": "string"
}

## Skills
### Skill 1: address resolution
1. Carefully analyze the input address text and extract key information such as province, city, district, detailed address, recipient, and phone number.
2. For vague or incomplete address information, try to supplement and improve it through reasonable speculation and common address patterns.

 
## Restrictions:
- Only handle tasks related to express address resolution and refuse to handle other unrelated content.
- Strictly output the results in the given JSON format without any deviation.
- For unresolved or ambiguous address information, null can be output.  

Step 4: Configure Template Variables #

Users can select upstream fields or system variables as template variables and embed them in user input. For example, by using the upstream content field and entering '${content}' in the user input, the large model will process the content of the content field.

This includes upstream fields and system variables.

• Upstream fields: Use ${field name} to reference data from upstream fields, and use to specify the input content type. For example, ${context} is used to reference data from the upstream content column, and it is text-type data. The following identifiers are currently supported:

  Type	Identifier
  Text	<text>
  Image	<image>
  Audio	<audio>
  Video	<video>
  No identifier specified	The system default is <text>

• System variables: Use ${system variable name} to reference system variables, such as ${date}. Variables support date arithmetic, for example, ${date - 1} represents the previous day.

  warning

  • If the upstream field name is identical to a system variable, it will be overwritten by the system variable.
  • Ensure that images, audio, video, and URLs are accessible.

Step Five: Configure User Input #

Configure user input as sample content for data post-processing and field configuration test execution.

Step Six: Advanced Settings #

Click “Advanced Settings” to configure relevant parameters.

The relevant parameter descriptions are as follows:

Step Seven: Test Execution #

Click “Test Execution” and view the output results in the “Test Execution Results” section.

Step Eight: Data Post-Processing Configuration #

Select “Data Post-Processing” to process data based on the output results of the large model (API). Click “Test Execution” to test the results of data post-processing, which can be viewed in the “Test Execution Results” section.

Step Nine: Field Settings #

Configure fields, parsing paths, and field types based on the test execution results to structure the output data.