Merge pull request #2750 from voutilad/ft-ptum-pup

Announce public preview of PTU-M for fine-tuned models.

Author: prmerger-automator[bot]

articles/ai-services/openai/how-to/fine-tuning.md

@@ -44,19 +44,98 @@ We use LoRA, or low rank approximation, to fine-tune models in a way that reduce
 
 ::: zone-end
 
-## Global Standard (preview)
+## Other Deployment Types
 
 Azure OpenAI fine-tuning supports [global standard deployments](./deployment-types.md#global-standard) in East US2, North Central US, and Sweden Central for:
 
 - `gpt-4o-mini-2024-07-18`
 - `gpt-4o-2024-08-06`
 
+And supports [regional provisioned managed](./deployment-types.md#provisioned) in North Central US and Switzerland West for:
+
+- `gpt-4o-mini-2024-07-18`
+- `gpt-4o-2024-08-06`
+
+### Global Standard (preview)
+
 Global standard fine-tuned deployments offer [cost savings](https://azure.microsoft.com/pricing/details/cognitive-services/openai-service/), but custom model weights may temporarily be stored outside the geography of your Azure OpenAI resource.
 
 :::image type="content" source="../media/fine-tuning/global-standard.png" alt-text="Screenshot of the global standard deployment user experience with a fine-tuned model." lightbox="../media/fine-tuning/global-standard.png":::
 
 Global Standard fine-tuning deployments currently do not support vision and structured outputs.
 
+### Provisioned Managed (preview)
+
+Provisioned managed fine-tuned deployments offer [predictable performance](../concepts/provisioned-throughput.md#what-do-the-provisioned-deployment-types-provide) for fine-tuned deployments. As part of public preview, provisioned managed deployments may be created regionally via the data-plane [REST API](../reference.md#data-plane-inference) version `2024-10-01` or newer. See below for examples.
+
+Provisioned Managed fine-tuning deployments currently do not support vision and structured outputs.
+
+#### Creating a Provisioned Managed deployment
+
+To create a new deployment, make an HTTP PUT call via the [Deployments - Create or Update REST API](/rest/api/aiservices/accountmanagement/deployments/create-or-update?view=rest-aiservices-accountmanagement-2024-10-01&tabs=HTTP&preserve-view=true). The approach is similar to performing [cross region deployment](#cross-region-deployment) with the following exceptions:
+
+- You must provide a sku name of `ProvisionedStandard`.
+- The capacity must be declared in PTUs.
+- The `api-version` must be `2024-10-01` or newer.
+- The HTTP method should be `PUT`.
+
+For example, to deploy a gpt-4o-mini model:
+
+```bash
+curl -X PUT "https://management.azure.com/subscriptions/<SUBSCRIPTION>/resourceGroups/<RESOURCE_GROUP>/providers/Microsoft.CognitiveServices/accounts/<RESOURCE_NAME>/deployments/<MODEL_DEPLOYMENT_NAME>api-version=2024-10-01" \
+  -H "Authorization: Bearer <TOKEN>" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "sku": {"name": "ProvisionedStandard", "capacity": 25},
+    "properties": {
+        "model": {
+            "format": "OpenAI",
+            "name": "gpt-4omini-ft-model-name",
+            "version": "1",
+            "source": "/subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/{SourceResourceGroupName}/providers/Microsoft.CognitiveServices/accounts/{SourceAOAIAccountName}"
+        }
+    }
+  }'
+```
+
+#### Scaling a fine-tuned model on Provisioned Managed
+
+To scale a fine-tuned provision managed deployment to increase or decrease PTU capacity, perform the same `PUT` REST API call as you did when [creating the deployment](#creating-a-provisioned-managed-deployment) and provide an updated `capacity` value for the `sku`. Keep in mind, provisioned deployments must scale in [minimum increments](../concepts/provisioned-throughput.md#how-much-throughput-per-ptu-you-get-for-each-model).
+
+For example, to scale the model deployed in the previous section from 25 to 40 PTU, make another `PUT` call and increase the capacity:
+
+```bash
+curl -X PUT "https://management.azure.com/subscriptions/<SUBSCRIPTION>/resourceGroups/<RESOURCE_GROUP>/providers/Microsoft.CognitiveServices/accounts/<RESOURCE_NAME>/deployments/<MODEL_DEPLOYMENT_NAME>api-version=2024-10-01" \
+  -H "Authorization: Bearer <TOKEN>" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "sku": {"name": "ProvisionedStandard", "capacity": 40},
+    "properties": {
+        "model": {
+            "format": "OpenAI",
+            "name": "gpt-4omini-ft-model-name",
+            "version": "1",
+            "source": "/subscriptions/00000000-0000-0000-0000-000000000000/resourceGroups/{SourceResourceGroupName}/providers/Microsoft.CognitiveServices/accounts/{SourceAOAIAccountName}"
+        }
+    }
+  }'
+```
+
+#### Deleting a Provisioned Managed deployment
+
+To delete a deployment, use the [Deployments - Delete REST API(/rest/api/aiservices/accountmanagement/deployments/delete?view=rest-aiservices-accountmanagement-2024-10-01&tabs=HTTP&preserve-view=true) and send an HTTP DELETE to the deployment resource. Like with creating deployments, you must include the following parameters:
+
+- Azure subscription id
+- Azure resource group name
+- Azure OpenAI resource name
+- Name of the PTU deployment to delete
+
+For example, to delete the model created and modified in the previous sections:
+
+```bash
+curl -X DELETE "https://management.azure.com/subscriptions/<SUBSCRIPTION>/resourceGroups/<RESOURCE_GROUP>/providers/Microsoft.CognitiveServices/accounts/<RESOURCE_NAME>/deployments/<MODEL_DEPLOYMENT_NAME>api-version=2024-10-01" \
+  -H "Authorization: Bearer <TOKEN>"
+```
 
 ## Vision fine-tuning