Customize base image for compute session

  • Article

This section assumes you have knowledge of Docker and Azure Machine Learning environments.

Step-1: Prepare the docker context

Create image_build folder

In your local environment, create a folder contains following files, the folder structure should look like this:

|--image_build
|  |--requirements.txt
|  |--Dockerfile
|  |--environment.yaml

Define your required packages in requirements.txt

Optional: Add packages in private pypi repository.

Using the following command to download your packages to local: pip wheel <package_name> --index-url=<private pypi> --wheel-dir <local path to save packages>

Open the requirements.txt file and add your extra packages and specific version in it. For example:

###### Requirements with Version Specifiers ######
langchain == 0.0.149        # Version Matching. Must be version 0.0.149
keyring >= 4.1.1            # Minimum version 4.1.1
coverage != 3.5             # Version Exclusion. Anything except version 3.5
Mopidy-Dirble ~= 1.1        # Compatible release. Same as >= 1.1, == 1.*
<path_to_local_package>     # reference to local pip wheel package

For more information about structuring the requirements.txt file, see Requirements file format in the pip documentation.

Define the Dockerfile

Create a Dockerfile and add the following content, then save the file:

FROM <Base_image>
COPY ./* ./
RUN pip install -r requirements.txt

Note

This docker image should be built from prompt flow base image that is mcr.microsoft.com/azureml/promptflow/promptflow-runtime:<newest_version>. If possible use the latest version of the base image.

Step 2: Create custom Azure Machine Learning environment

Define your environment in environment.yaml

In your local compute, you can use the CLI (v2) to create a customized environment based on your docker image.

Note

az login # if not already authenticated

az account set --subscription <subscription ID>
az configure --defaults workspace=<Azure Machine Learning workspace name> group=<resource group>

Open the environment.yaml file and add the following content. Replace the <environment_name> placeholder with your desired environment name.

$schema: https://azuremlschemas.azureedge.net/latest/environment.schema.json
name: <environment_name>
build:
  path: .

Create an environment

cd image_build
az ml environment create -f environment.yaml --subscription <sub-id> -g <resource-group> -w <workspace>

Note

Building the environment image may take several minutes.

Go to your workspace UI page, then go to the environment page, and locate the custom environment you created.

You can also find the image in environment detail page and use it as base image for compute session of prompt flow. This image will also be used to build environment for flow deployment from UI. Learn more how to specify base image in compute session.

To learn more about environment CLI, see Manage environments.