Edit

Share via

Fine-tune AI models in Microsoft Foundry with the Azure Developer CLI fine-tuning extension

In this article, you learn to use the Azure Developer CLI (azd) AI fine-tuning extension to set up and run fine-tuning jobs in Microsoft Foundry. The extension lets you initialize projects from templates, submit and manage fine-tuning jobs, and deploy fine-tuned models directly from your terminal.

Prerequisites

  • The Azure Developer CLI (azd) installed (version 1.22.1 or later) and authenticated (azd auth login).
  • The azd AI fine-tuning extension installed (azd ext install azure.ai.finetune). See Install the fine-tuning extension for details.
  • An Azure subscription with permission to create and manage Microsoft Foundry resources.
  • (Optional) The GitHub CLI installed, if you plan to download sample templates from GitHub repositories.

Install the Azure Developer CLI

Install azd for your operating system.

winget install microsoft.azd

Verify the installation

After installing, verify that azd is installed and meets the minimum version requirement:

azd version

Confirm the output shows version 1.22.1 or later. If you need to upgrade, run:

winget upgrade Microsoft.azd

Install the fine-tuning extension

Add the Azure AI fine-tuning extension to azd:

azd ext install azure.ai.finetune

Verify the extension is installed:

azd ext list

Authenticate

Sign in to Azure (required)

Authenticate with your Azure account to access your subscription and resources:

azd auth login

Initialize your project

Use the azd ai finetuning init command to scaffold a fine-tuning project. Navigate to your desired working directory before running any of the following initialization methods.

Tip

You can skip initialization entirely by using the Quick Submit option, which lets you submit a fine-tuning job by providing the subscription and Foundry project endpoint inline.

Find your project ARM resource ID

To initialize with an Azure AI Foundry project, you need the project's ARM resource ID. The resource ID follows this format:

/subscriptions/{sub}/resourceGroups/{rg}/providers/Microsoft.CognitiveServices/accounts/{account}/projects/{project}

You can find this value in the Azure portal by navigating to your AI Foundry project's Profile page under Project details.

Option 1: Project + Template

Use an existing Azure AI Foundry project with a template:

azd ai finetuning init -p <project-resource-id> -t <template-url>

Example:

azd ai finetuning init \
  -t https://github.com/achauhan-scc/foundry-samples/blob/main/samples/python/finetuning/supervised \
  -p /subscriptions/{sub}/resourceGroups/{rg}/providers/Microsoft.CognitiveServices/accounts/{account}/projects/{project}

Option 2: Project + Existing job

Clone configuration from an existing fine-tuning job:

azd ai finetuning init -p <project-resource-id> -j <job-id>

Example:

azd ai finetuning init \
  -p /subscriptions/{sub}/resourceGroups/{rg}/providers/Microsoft.CognitiveServices/accounts/{account}/projects/{project} \
  -j ftjob-4cad7de198a34baeb4f0c95ff01ac844

Option 3: Template only

Start from a template and configure the project later:

azd ai finetuning init -t <template-url>

Example:

azd ai finetuning init -t https://github.com/achauhan-scc/foundry-samples/blob/main/samples/python/finetuning/supervised

Option 4: Clone from job

Clone configuration from an existing job ID:

azd ai finetuning init -j <job-id>

Example:

azd ai finetuning init -j ftjob-4cad7de198a34baeb4f0c95ff01ac844

Option 5: Project endpoint only

Initialize with just your Azure AI Foundry project endpoint:

azd ai finetuning init -e <project-endpoint>

Example:

azd ai finetuning init -e https://account.services.ai.azure.com/api/projects/project-name

Option 6: Interactive mode

Run without parameters for guided setup prompts:

azd ai finetuning init

Option 7: Minimal init (with subscription lookup)

Use minimal initialization for a simplified interactive experience with subscription lookup:

azd init --minimal

This option provides guided prompts to select your subscription and configure your environment.

Run fine-tuning commands

Navigate to your project folder (where fine-tune-job.yaml is located) and use the following commands to manage fine-tuning jobs.

Tip

Looking for example job YAML files? Check out the Fine-tuning CLI Samples in the Foundry samples repository.

Quick submit (skip initialization)

You can submit a job directly without running azd init first by providing the subscription and project endpoint inline:

azd ai finetuning jobs submit -f <path-to-yaml> -s <subscription-id> -e <project-endpoint>

Example:

azd ai finetuning jobs submit \
  -f /path-from-working-directory-to-config/job.yaml \
  -s a9096eb7-bfec-47e8-be27-b040b82afac9 \
  -e https://my-resource.services.ai.azure.com/api/projects/my-project
Parameter Description
-f Path to the job YAML file
-s Azure subscription ID
-e Project endpoint URL

Submit a job

azd ai finetuning jobs submit -f ./fine-tune-job.yaml

List jobs

azd ai finetuning jobs list

Show job details

azd ai finetuning jobs show -i <job-id>

Pause a job

azd ai finetuning jobs pause -i <job-id>

Resume a job

azd ai finetuning jobs resume -i <job-id>

Cancel a job

azd ai finetuning jobs cancel -i <job-id>

Deploy your fine-tuned model

Once your fine-tuning job completes successfully, deploy the model for inference:

azd ai finetuning jobs deploy -i <job-id> -d "<deployment-name>" -c 100 -m "OpenAI" -s "GlobalStandard" -v "1"
Parameter Description
-i Job ID
-d Deployment name
-c Capacity
-m Model provider
-s SKU name
-v Version

Quick reference

Init parameters

Parameter Description
-p Project resource ID (ARM)
-e Project endpoint URL
-t Template URL or path
-j Clone from job ID
-w Working directory
-n Environment name
-s Subscription ID

Job parameters

Parameter Description
-f YAML file path
-i Job ID

Additional resources

  • Last updated on