Manage tasks

Available in Classic and VPC

Task management describes how to create and manage tasks in the Tuning menu and how to check the learning status. It also introduces how to use Inference Test to evaluate the performance and accuracy of trained tasks and how to create a test app.

Create a new task

You can test a part of a pre-trained model as retraining it to your dataset by creating a new job.

The following describes how to create a new task.

1. In the NAVER Cloud Platform console, click the Services > AI Services > CLOVA Studio JP menus, in that order.
2. Click the My Product menu.
3. Click the [Go to CLOVA Studio JP] button.
4. Click the Tuning menu.
5. Click the [Create] button for the type you want to work on.
   • Tuning provides a total of 7 types, and you can check the description and application examples of the task types in the form of cards.
6. When you see the pop-up window, click the [Create] button on selecting the model engine.
   • You cannot change the model engine later.
   • For a more detailed description of the model engine, see Engine.
7. Enter a task name.
8. Please upload datasets after clicking on the file upload area and checking the personal information and harmful information guide.
   Caution

   Be sure to upload the dataset file in compliance with its specifications to ensure normal learning progress and performance. See Dataset for more details.

   • For .csv files, you can download the dataset format file by clicking Download format.
   • If you upload the dataset successfully, you can see the dataset file name and size in the file upload area.
   • Even if you upload the dataset successfully, when improvements are recommended for optimal performance, you can view the caveats by clicking Check cautions.
   • If you fail to upload the dataset, you can see the file name and reason for failure in the file upload area. You can check specific errors on clicking View details.
9. Click the [Next] button.
10. Check the expected token usage and click the [Learn] button in the Token Calculation pop-up window.
11. Check the contents of the pop-up window while learning is waiting, and then click the [OK] button.
    • You can check Learning status in the [Username] > [My Task] > Tuning tab.

Dataset

Please check the specifications, precautions, and examples for dataset files and create the dataset files correctly.

Dataset file specifications and precautions

Dataset file specifications and precautions are as follows.

Common

• Enter the a file name between 2 to 30 characters.
• Only .csv and .jsonl file extensions are supported.
• The encoding format of the file supports UTF-8 format.
• You can only upload files with a file size of 50 MB or less.
• Please enter at least 1000 valid data.
• For document classification tasks, the number of datasets per category should be equal, and we recommend a minimum of 200 cases. In addition, we recommend configuring a classification label by a single word excluding spacing and special characters.
  • <example> 30% positive (300 cases), 30% negative (300 cases), 40% neutral (400 cases)
• When you need to break the line, please use ‘\n.’
• You are responsible for any problems arising from uploading datasets containing personal information.

.csv file

• The first row must accurately contain "Text" and "Completion" and it must consist of only two columns.
• Be sure to delete blank rows and columns.
• Each row (pair of texts and completions) of the file must not exceed 1000 characters, including spaces. If it exceeds, only a part of the dataset is uploaded.

.jsonl file

• Each line must consist of {"Text": "Input value", "Completion": "Desired result"}, and "Input value" and "Desired result" must contain at least one character.
• Please enclose double quotation marks with '"'.

Example of composing a dataset file

Composing examples of a dataset file is as follows:

• For dialog tasks, please fill out the dataset as follows to ensure optimal performance.
  • Input 3 or more sentences into the Text column, and input 1 sentence into the Completion column.
  • Unify the speaker of the Completion column to one person
  • We recommend configuring so that the speeches are continued between the Text column and the Completion column.
  • The agents of speech (speakers) are limited to 2 people.
  • Make sure to specify the subject of speech before the statement (<example> 'Customer:', 'Seller:')

Check learning status

The following describes how to check the learning status.

1. In the NAVER Cloud Platform console, click the Services > AI Services > CLOVA Studio JP menus, in that order.
2. Click the My Product menu.
3. Click the [Go to CLOVA Studio JP] button.
4. After selecting [User Name] in the upper right corner, click [My Task] > Tuning.
   
   • Create a new task: Click to move to the [New Task] tab
   • /: Click to change sorting
   • Waiting for learning: Waiting for learning, you can click to see a pop-up window of waiting for learning
   • Learn in progress: Learning is in progress, you can click to check the estimated time required
   • Learning completed: Learning is complete, you can click to check and test task information
   • Suspend learning: The state of stopping learning

Check task information to complete learning

Click a completed task to view its information.

• Creation date and time: Date and time of creation of a new task
• Learning completion date: Date and time of completion of learning
• Workflow ID: An ID for identifying a task in learning
• Problem Type: Task type
• Model Engine: Type of learned language model
• Dataset: Dataset file name used for learning
• Train Loss: A figure that shows how proper the model is to the dataset. The lower the figure is, the less the error from the correct answer.
• Tokens Used: Number of tokens actually used

Suspend learning

The following describes how to suspend learning.

1. In the NAVER Cloud Platform console, click the Services > AI Services > CLOVA Studio JP menus, in that order.
2. Click the My Product menu.
3. Click the [Go to CLOVA Studio JP] button.
4. After selecting [User Name] in the upper right corner, click [My Task] > Tuning.
5. Please click on the task to stop learning.
6. Click the [Stop] button.
7. Check the contents of the pop-up window of suspending learning, and then click the [Stop] button.
   • If you stop the pending training, you will lost all previously uploaded dataset files.
   • If you stop training in progress, you will lost previously uploaded dataset files and jobs, and the token that was notified when creating the job will be used.

Utilize trained tasks

Provides an Inference test function to evaluate performance and accuracy before creating a test app for a trained task. After completing the test, you can create a test app, and you can also share your work via a share URL.

Inference Test

The following describes how to test with the Inference Test function.

1. In the NAVER Cloud Platform console, click the Services > AI Services > CLOVA Studio JP menus, in that order.
2. Click the My Product menu.
3. Click the [Go to CLOVA Studio JP] button.
4. After selecting [User Name] in the upper right corner, click [My Task] > Tuning.
5. Among the trained tasks, click the desired task to test.
6. After entering the input value in the Input area, click the [Run] button.

• You can enter up to 2000 characters including spaces.
• You can see the completion generated through learning in the Output area.
• Please verify performance and accuracy through sufficient testing.

Create test app

The following describes how to create the test app.

1. In the NAVER Cloud Platform console, click the Services > AI Services > CLOVA Studio JP menus, in that order.
2. Click the My Product menu.
3. Click the [Go to CLOVA Studio JP] button.
4. After selecting [User Name] in the upper right corner, click [My Task] > Tuning.
5. Click the task you want to create a test app for.
6. Click the [Create test app] button.
7. Enter a test app name, and then click the [Create] button.

   • You can create a test app, and see a test app pop-up window.

     

     • You can check the API information of the test app, and you can set whether to use AI Filter. (See CLOVA Studio JP API Guide for details on the API.)
     • As for the code type, curl and python are provided.
     • You can click [Copy] to copy the API information to the clipboard.
     • You can click [Reissue] to reissue API Gateway Key.
     • You can click View guide to check the AI Filter guide.