Manage tasks on tuning

Prev Next

Available in Classic and VPC

Explains how to train models, create, and manage tasks using the tuning API. It also introduces how to use inference test to evaluate the performance and accuracy of completed training tasks.

Start tuning API

You can test a part of a pre-trained model by retraining it to your dataset by creating a new tuning task. You need a dataset that meets upload requirements for tuning tasks. The larger the dataset, the better the tuning results.

Use the tuning API by calling it directly. For instructions on calling the tuning API, see the API guides.

Caution
  • The greater the number of tokens in the dataset, the longer it takes to tune and more cost may be incurred.
  • Prior to training, it may take up to 6 hours to wait for training to secure a GPU for training and pre-process the data.
Note
  • Training creation and retrieval can be called according to the API specification provided on the service page.
  • To generate results from the trained model, use the chat completions API or v3 chat completions API.

Check training status

To check the training status, follow these steps:

  1. From the NAVER Cloud Platform console, click i_menu > Services > AI Services > CLOVA Studio in order.
  2. Click the My product menu, and then click the [Go to CLOVA Studio] button.
  3. On the left side of the screen, click the My tasks menu.
  4. Click the [Tuning] tab.
    clovastudio-tuning01_tasklist_ko
    • Training completed: training is complete, you can click to check and test task information.
    • Training pending: waiting for training, you can click to see a popup window of waiting for training.
    • Training in progress: training is in progress, you can click to check the run time.
    • Training failed: training was not successfully performed.
    • Suspend training: training is suspended.

Check completed training tasks

Click a training-completed task to view its information.

  1. From the NAVER Cloud Platform console, click i_menu > Services > AI Services > CLOVA Studio in order.
  2. Click the My product menu, and then click the [Go to CLOVA Studio] button.
  3. On the left side of the screen, click the My tasks menu.
  4. Click the [Tuning] tab.
  5. Click the task marked as 'Training completed' in the list.
  6. Check detailed task information on the details page.
    • Created: date and time the task was created
    • yyyy-mm-dd: date and time training was completed
    • Workflow ID: an ID for identifying a task in training
    • Problem type: task type
    • Model: type of trained language model
    • Dataset: dataset file name used for training
    • 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
Note
  • You can also check your training status when you are using the API call.
  • The number of tokens displayed when training is interrupted may differ from the number of tokens displayed in NAVER Cloud Platform console.

Suspend training

Note

You can suspend only tasks waiting for training or training in progress.

To suspend training, follow these steps:

  1. From the NAVER Cloud Platform console, click i_menu > Services > AI Services > CLOVA Studio in order.
  2. Click the My product menu, and then click the [Go to CLOVA Studio] button.
  3. On the left side of the screen, click the My tasks menu.
  4. Click the [Tuning] tab.
  5. Click on the task to suspend training.
  6. Click the [Suspend] button.
  7. Check the contents of the popup window of suspending training, and then click the [Suspend] button.
    • If you suspend a pending training, you lose all previously uploaded dataset files.
    • If you suspend training in progress, you will lose previously uploaded dataset files and tasks, and the token that was notified when creating the task will be used.
Caution
  • If you suspend training a task in progress, you may be charged usage fees for used tokens depending on the training progress.
  • You cannot resume the suspended training.
  • You can suspend training in the same way when you are using the API.

Using training completed job

You can perform an inference test on completed tasks to check performance and accuracy in advance. Based on the test results, you can generate a URL to share the task.

To test with the inference test feature, follow these steps:

  1. From the NAVER Cloud Platform console, click i_menu > Services > AI Services > CLOVA Studio in order.
  2. Click the My product menu, and then click the [Go to CLOVA Studio] button.
  3. On the left side of the screen, click the My tasks menu.
  4. Click the [Tuning] tab.
  5. Click the task to test.
    • You can only test tasks that are in the training-completed status.
  6. Go to playground to run the test.