Creating a Data Instruction

Description

Data instructions consist of user-defined input variables (such as parameters with the {{}} symbol) and prompt. They are used to automatically generate or process data based on user requirements.

Currently, data instructions can be used in data synthesis tasks and support only text datasets.

Procedure

This section uses the scenario of generating a topic prose as an example to describe how to configure a data synthesis instruction.

1. In the Create Instruction dialog box, configure an instruction.
   1. Enter variables in double curly brackets {{}}.
   2. Enter the following prompt: "Please write an essay with no more than {{num}} words on the theme of {{topic}}."
   3. Click Identify and then click Yes.
2. Set variables based on Table 1.

   Table 1 Data instruction variable configuration

   Type	Variable Name	Variable Type	Variable Description
   Input variables	topic	string	Topic
   	num	string	Word count
   Output variables	output	string	Prose

   The variable description field of an output variable is the content that can be understood by the large model and needs to be filled in carefully.

3. Commission the instruction.
   • In the Commissioning > Model area, select the required model, and customize the hyperparameter values.
   • In the Commissioning > Input area, check the model effectiveness by assigning values to variables. For example, set topic to autumn and num to 100.
4. Click Generate to view the generated result. After the commissioning is complete, click Create Now to create the data instruction.