2.4

Add InfuseAI Image

InfuseAI provides and maintain a serious of public images on DockerHub. This document shows how to add images provided by InfuseAI.

First of all we check and pick up a pari of images from Avaialbe Image List.

They are TensorFlow 2.1/CPU/Python3.7 and TensorFlow 2.1/GPU CUDA 10.1/Python3.7.

We can learn URLs respectively from the list.

  • TensorFlow 2.1/CPU/Python 3.7: infuseai/docker-stacks:tensorflow-notebook-tf-v2-3f48358e

  • TensorFlow 2.1/GPU CUDA10.1/Python 3.7: infuseai/docker-stacks:tensorflow-notebook-3f48358e-gpu-cuda-10

Steps

  1. Login as an administrator, enter Admin Dashboard and Images Management.

  2. Click + Add.

  3. Fill in Name: such as tf-21.

  4. Fill in Description: such as TensorFlow 2.1 + Cuda10.1 + Python3.7.

    1. Select Type: Universal。

      In terms of Universal, we can assign a CPU-only image and a GPU-supported image respectively. When spawning a JupyterHub, PrimeHub will pick up the corresponding image according to the specified instance type (CPU-only or GPU-required).

  1. Fill in Container image url: CPU-only infuseai/docker-stacks:tensorflow-notebook-tf-v2-3f48358e

  2. Check off Specific container image url for GPU and fill in GPU-supportedinfuseai/docker-stacks:tensorflow-notebook-3f48358e-gpu-cuda-10.

  3. Because InfuseAI's registry on DockerHub is public, there is no need of Use Image Pull Secret.

  4. Global or specify accessing Groups.

Verify

  1. Enter JupyterHub from User Portal.

  2. Select a group which can utilize the newly added image.

  3. Select an instance type which requires small CPU resource. (just for verification)

  4. Select the newly added Image tf-21.

  5. Start Notebook. It takes time to pull off the newly added image for the first time; As long as a JupyterHub is launched successfully, the image is pulled successfully.

Verify GPU Image (Please avoid affecting users )：

  1. Close/Stop previously launched JupyterHub. REF

  2. Enter JupyterHub from User Portal.

  3. Select a group which can utilize the newly added image.

  4. Select an instance type which requires small GPU resource. (just for verification)

  5. Select the newly added Image tf-21.

  6. Start Notebook. It takes time to pull off the newly added image for the first time; As long as a JupyterHub is launched successfully, the image is pulled successfully.

← Build ImageAdd Image →