Tensorboards

Available in VPC

This section describes the Tensorboards interface. Use Tensorboards to create and delete Tensorboards or access active Tensorboards to monitor training processes.

Tensorboards

The Tensorboards interface includes the following components:

View Tensorboards list

The Tensorboards list includes the following information:

• Status: Tensorboard status
• Name: Tensorboard name
• Created At: Initial creation date and time.
• LogsPath: Storage path for logs
• CONNECT: Click to open the Tensorboard in a new window

Create a Tensorboard

To create a new Tensorboard:

1. Click [Create].
2. When the New Tensorboard popup appears, enter a Tensorboard name.
   • Enter 3-61 characters using lowercase letters (a-z), numbers (0-9), and hyphens (-).
   • It must start with an English letter and end with an English letter or a number.
   • Duplicate names are not allowed.
   • Must be unique across Tensorboards and existing Notebooks.
3. Verify the Project Namespace.
4. Select a storage type for log storage.
   • Object Storage/Ncloud Storage: Use your Object Storage or Ncloud Storage
   • Volumes: Use a volume you created
5. Enter the Mount Path for the selected storage type.

6. Select the configuration for the chosen Storage Type.
   • If you have chosen Object Storage or Ncloud Storage, select the PodDefault created in advance.
   • If you have chosen Volumes, select the GPU Zone assigned to your Workspace. For GPU Zone information, see View available GPU Zones.
7. Click [Create].

Delete a Tensorboard

You can delete a Tensorboard as follows:

1. Select the Tensorboard you want to delete and click [Delete].
2. When the Delete Tensorboard popup appears, enter the Tensorboard name and click [Delete].
3. Verify that the selected Tensorboard has been removed from the list.

Configure integration settings by storage type

You can configure Object Storage and Ncloud Storage on NAVER Cloud Platform for integration with Tensorboard. After creating a Tensorboard, follow the steps below in order.

1. Check settings by storage type

Object Storage and Ncloud Storage both provide S3-compatible interfaces and can be integrated with Tensorboard. Because endpoint and region settings differ by storage type, refer to the table below when configuring them. Here are the differences in settings by storage type:

2. Create a Secret

Create an appropriate Credential Secret for the storage type you want to use.

Object Storage

kubectl create secret generic ncp-object-storage-credentials \
  -n <your-namespace> \
  --from-literal=AWS_ACCESS_KEY_ID=<your-access-key> \
  --from-literal=AWS_SECRET_ACCESS_KEY=<your-secret-key>

Ncloud Storage

kubectl create secret generic ncp-ncloud-storage-credentials \
  -n <your-namespace> \
  --from-literal=AWS_ACCESS_KEY_ID=<your-access-key> \
  --from-literal=AWS_SECRET_ACCESS_KEY=<your-secret-key>

3. Create a PodDefault

Create an appropriate PodDefault for the storage type you want to use.

Object Storage

For Object Storage, the region setting is required. Set it to kr-standard.

apiVersion: kubeflow.org/v1alpha1
kind: PodDefault
metadata:
  name: tensorboard-ncp-object-storage-config
spec:
  desc: NCP Object Storage config for TensorBoard
  selector:
    matchLabels:
      tb-ncp-object-storage-config: "true"
  env:
    - name: AWS_ACCESS_KEY_ID
      valueFrom:
        secretKeyRef:
          name: ncp-object-storage-credentials
          key: AWS_ACCESS_KEY_ID
    - name: AWS_SECRET_ACCESS_KEY
      valueFrom:
        secretKeyRef:
          name: ncp-object-storage-credentials
          key: AWS_SECRET_ACCESS_KEY
    - name: AWS_REGION
      value: kr-standard
    - name: S3_ENDPOINT
      value: https://kr.object.ncloudstorage.com

Ncloud Storage

apiVersion: kubeflow.org/v1alpha1
kind: PodDefault
metadata:
  name: tensorboard-ncp-ncloud-storage-config
spec:
  desc: NCP NCloud Storage config for TensorBoard
  selector:
    matchLabels:
      tb-ncp-ncloud-storage-config: "true"
  env:
    - name: AWS_ACCESS_KEY_ID
      valueFrom:
        secretKeyRef:
          name: ncp-ncloud-storage-credentials
          key: AWS_ACCESS_KEY_ID
    - name: AWS_SECRET_ACCESS_KEY
      valueFrom:
        secretKeyRef:
          name: ncp-ncloud-storage-credentials
          key: AWS_SECRET_ACCESS_KEY
    - name: S3_ENDPOINT
      value: https://kr.ncloudstorage.com