Skip to Content
Information

This site will undergo a brief period of maintenance on Thursday, 23 April at 2:30 AM Central/1:00 PM IST. During a 30 minute window, site availability may be intermittent.

Configuring settings to use a fine-tuned model to generate Best action recommendations for Ticket Resolver

Information

Important

This topic is applicable only for BMC Helix ITSM SaaS subscribers.

BMC Helix ITSM connects with BMC HelixGPT, a generative AI capability available for BMC Helix applications, to use autonomous agents for operators or SREs who can use them to get AI-driven insights for managing complex IT environments.

A fine-tuned model is available for using agentic AI capabilities in BMC Helix ITSM. As a tenant administrator, you deploy the fine-tuned model in one of the supported cloud platforms.

Information

Important

You need ITSM components (BMC Helix Innovation Studio and BMC Helix ITSM Insights) to configure model settings and agents in BMC Helix Agent Studio. These components are available as part of the BMC Helix ITSM CORE version. To obtain these components, contact BMC Helix Support.

Before you begin

Perform the following steps before deploying the fine-tuned model in your cloud:

Warning
Important

These are the minimum requirements to deploy the fine-tuned model in your cloud environment. For details, see the documentation for your selected cloud.

The process to deploying the model remains the same, regardless of your license entitlements. 

Google Cloud Platform requirements:

  • You have an active Google Cloud Platform subscription and a GCP project in a Vertex AI-supported region.
    All resources and artifacts must be kept in the same region. 
  • You have the Identity and Access Management (IAM) permissions to perform the following tasks:

    • Write to the target Google Cloud Storage bucket (roles/storage.admin).

    • Register and deploy models: Vertex AI Administrator (roles/aiplatform.user) or equivalent role.

    • Access the Artifact Registry or Container Registry (if using custom containers stored in GCP).

Local host requirements:

  • Google Cloud SDK is installed, and the project where you want to deploy the model is set.
  • The gsutil tool is available. You need this tool to upload the model artifacts to the Google Cloud Storage Bucket. 
  • ​​​​Docker Engine is installed and running.

Microsoft Azure AI requirements:

  • You have an active Microsoft Azure subscription. 
  • You must have the following roles and permissions in Microsoft Azure:
    • Contributor OR ML Data Scientist and Compute Operator
    • Storage Blob Data Contributor
    • AcrPull/AcrPush
    • KeyVault Secrets User
    • Application Insights Contributor
    • Log Analytics Contributor ​​​​​​
  • You have the Standard_NC24ads_A100_v4 or Standard NCADSA100v4 quota assigned to your region:
    1. In the Azure Machine Learning studio, click Quota.
    2. Click the subscription name for the subscription where you want to host the model.
    3. Select the region.
    4. Search and select Standard NCADSA100v4 Family Cluster Dedicated vCPUs.
    5. Click Request quota.
       If the unused quota is 0 or less than 24, click Request quota and set the New cores limit to whatever the Usage is plus 24. So if Usage is 0, set Quota to 24. This quota is enough for a machine type with one A100 accelerator.  
    6. Click Submit.
      After your quota limit is approved, you must assign it to the workspace later. 

Local host requirements:

  • The Microsoft Azure CLI is installed with the Azure Machine Learning (ml) extension.
  • Docker Engine is installed and running.

Task 1: To obtain a model from BMC Helix

BMC Helix requires you to contact BMC Helix support to obtain the latest fine-tuned model for AIOps from BMC Helix. You may select from the following fine-tuned foundation models:

  • QWEN

During the term of the customer's license, you may request a change of the fine-tuned foundation model by contacting BMC Helix support. 

Note that BMC Helix provides a Docker image tarball file with all model artifacts.

Task 2: To deploy the model in your cloud

Based on your cloud platform, perform the steps to deploy the fine-tuned model.  

These are the minimum steps required to deploy the fine-tuned model in your cloud environment. For details, see the documentation for your selected cloud. 

You import a model into the model registry and associate it with a container. From the model registry, you can deploy your imported model to an endpoint.

To import the model

  1. On a local host, extract the model artifacts provided by BMC Helix: 
    tar -xzvf <helix_gpt_model_version>.tar.gz
  2. Upload the model to the Google Cloud Storage bucket:
    gsutil cp -r <helix_gpt_model_version> gs://<your-bucket>/model/
  3. Prepare the Custom Inference Docker image by loading it on your host:
    docker load -i /path/to/model_container.tar​​​​​​
  4. Push the Docker image to the Google Cloud container registry: 
    #docker tag <bmc helix image> <Google Cloud Container Registry tag>

    Example: docker tag helix-gpt-inference:<version_number> \gcr.io/my-gcp-project/helix-gpt-inference:<version_number>

    #docker push <Google Cloud Container Registry path>

    Example: docker push gcr.io/my-gcp-project/helix-gpt-inference:<version_number>
    Now, the model and its artifacts are available in the Google Cloud Model Store and the Google Cloud Container Registry.
  5. Navigate to Model Registry from the Vertex AI navigation menu.
  6. Click Import and then click Import as new model.
  7. On the Import Model page, provide the name of the model, select the region, and click Continue.
    Select the region that matches both your bucket's region and the Vertex AI regional endpoint that you are using. 
  8. Navigate to the Model settings page and select Import an existing container.
  9. In the Custom container settings section, click Browse in the Container image field and then click the Container Registry tab to select the container image.
  10. Click Browse in the Model artifact location and select the Cloud Storage path to the directory that contains your model artifacts.
  11. In the Arguments section, specify the following parameters and click Continue:
    FieldDescriptionExample value
    Environment variables Specify the file name of the Deployment spec (without the file extension) included in the model artifacts.DEPLOYMENT_SPEC=

    zhp52uqvaxvacmt4u2tbezojfucjkf4f-helix-gpt-v6-instruct

    Prediction routeSpecify the HTTP path to send prediction requests to. /predictions
    Health routeSpecify the HTTP path to send health checks to./ping
    PortSpecify the port number to expose from the container. 8080
  12. On the Explainability options page, retain the default No explainability option, and click Import.
    After a few minutes, the model is displayed on the Models page.
    For more information about importing models in GCP Vertex AI, see the online documentation https://cloud.google.com/vertex-ai/docs/model-registry/import-model#custom-container.  

To deploy the model and create an endpoint

  1. Select the model and then click Deploy and test.
  2. Click Deploy to endpoint and then click Create new endpoint.
  3. Type the name of the endpoint and make sure that the region is the same as that of the model.
  4. Retain the access setting value to Standard and click Continue.
  5. On the Model Setting Page, specify the values for the following fields and continue with default values for other fields:
    • Machine Type: a2-highgpu-1g, 12 vCPUs, 85 GiB Memory
    • Accelerator Type: NVIDIA Tesla A100
    • Accelerator Count: 1
  6. Click Continue and then click Deploy.
    After the model is deployed, note the following information. These parameters are required when you configure the model in BMC HelixGPT Manager in the next step.
    FieldDescription
    IDContains the endpoint ID.
    Region

    The region where the model is deployed.

    For example, us-central1.

    Project ID

    Contains the project ID.

    For example, sso-gcp-dsom-sm-pub-cc39921.

To obtain the API key for Google

For Google, only the API Key method for authentication is supported. You need the service account API key and other details to configure the model in BMC HelixGPT Agent Studio in the next step. The service account must have the following Identity and Access Management (IAM) permissions:

  • aiplatform.endpoints.get
  • aiplatform.endpoints.predict

​​

Perform the following steps to obtain the API key for Google
  1. Log on to the Google Cloud Console with the same credentials you used while deploying the model.
  2. In the Admin account from the Main menu, select IAM & Admin >  Service Accounts.
  3. On the Service Accounts page, select the Keys tab and click Add keys, and then select Create New keys.
  4. Select the key type as JSON.
    The API key is downloaded.
    {
      "type": "service_account",
      "project_id": "redacted",
      "private_key_id": " redacted ",
      "private_key": redacted ",
      "    client_email": " redacted ",
      "    client_id": " redacted ",
      "    auth_uri": "https://accounts.google.com/o/oauth2/auth",
      "    token_uri": "https://oauth2.googleapis.com/token",
      "    auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs",
      "    client_x509_cert_url": " redacted ",
      "    universe_domain": "googleapis.com"     ​​​

 

You can deploy the model by using the Microsoft Azure Machine Learning Studio, however, the following steps explain how to use the Azure command line interface (CLI) for more control over the deployment.

Perform the following steps:

  1. On a local host, extract the model artifacts provided by BMC Helix: tar -xzvf <helix_gpt_model_version>.tar.gz
  2. Log in to the Microsoft Azure CLI: 
    az login

  3. Create a resource group:
    az group create --name <resource-group-name> --location <azure-region>

  4. Create an Azure Container Registry:az acr create

    --resource-group <resource-group-name>
    --name <name of the registry>
    --sku Basic
    --admin-enabled true
    ParameterDescription
    resource-groupSpecify the name of the resource group created in the previous step.
    nameSpecify the name of the Azure Container Registry.
    skuSpecify the pricing tier: Basic, Standard, or Premium. Most users start with Basic.
    adminSpecify true to make sure that the user can upload resources to the registry.

  5. Tag the docker image:
    docker tag <local-image>:<tag> helixgptreg.azurecr.io/vllm-vertex:<tag>

  6. Push the docker image to the Azure Container Registry:
    docker push helixgptreg.azurecr.io/vllm-vertex:<tag>
  7. Create an Azure ML workspace:
    az ml workspace create
      --name <workspace name> \
      --resource-group <resource-group-name> \
      --location <region> \
  8. Set Azure CLI default values:
    az configure --defaults workspace=<workspace name> group=<resource-group-name> location=<region>
  9. Link the Azure Container Registry to the workspace:
    az ml workspace update \
    --name <workspace-name> \
    --resource-group <resource-group-name>\
    --container-registry
    /subscriptions/73ae88b6-0681-4b5f-839d-e331f68a59ae/resourceGroups/<resource-group-name>/providers/Microsoft.ContainerRegistry/registries/helixgptreg \
    --update-dependent-resources
  10. Create an online endpoint:az ml online-endpoint create --file <endpoint>.yml
    endpoint.yaml:
    $schema: https://azuremlschemas.azureedge.net/latest/managedOnlineEndpoint.schema.json
    name: helix-gpt-v7-25-3-endpoint
    auth_mode: key
  11. Create an online deployment:
    az ml online-deployment create --file azure.yml --all-traffic
    Sample YAML file:
    Azure.yaml:
    name: helix-gpt-v7-25-3-deploy
    endpoint_name: helix-gpt-v7-25-3-endpoint
    model:
      name: helix-gpt-v7-25-3
      path: ./helix-gpt-v7-25-3
      version: 1
    environment_variables:
      AIP_HEALTH_ROUTE: "/ping"
      AIP_PREDICT_ROUTE: "/score"
      MODEL_BASE_PATH: "/var/azureml-app/azureml-models/helix-gpt-v7-25-3/1/helix-gpt-v7-25-3"
      DEPLOYMENT_SPEC: "agaqnayhu2tstm7s3z5xmnmdugrzccsa-helix-gpt-v7_2"
      AIP_STORAGE_URI: "/var/azureml-app/azureml-models/helix-gpt-v7-25-3/1/helix-gpt-v7-25-3"
    environment:
      image:attach:xwiki:Service-Management.IT-Service-Management.BMC-Helix-ITSM.itsm261.Administering.Configuring-settings-to-use-a-fine-tuned-model-to-generate-Best-action-recommendations-for-Ticket-Resolver.WebHome@filename helixgptreg.azurecr.io/vllm-vertex:dfe4802-43
      inference_config:
        liveness_route:
          port: 8080
          path: /ping
        readiness_route:
          port: 8080
          path: /ping
        scoring_route:
          port: 8080
          path: /score
    request_settings:
      request_timeout_ms: 180000
    instance_type: Standard_NC24ads_A100_v4
    instance_count: 1
  12. (_Optional_) Get container logs:
    az ml online-deployment get-logs \
    --endpoint-name <name of the endpoint> \
    --name <name of the deployment>
  13. Set traffic:
    When you set traffic to 100%, all requests sent to the endpoint are routed to that single deployment.
    az ml online-endpoint update \
    --name <name of the endpoint> \
    --resource-group <resource-group-name>\
    --workspace-name <workspace-name> \
    --traffic "<workspace-name>-deploy=100"
  14. Get the scoring URI:
    The scoring URI is the REST API endpoint you use to send data and get predictions from your deployed model.
    az ml online-endpoint show \
    --name <endpoint-name>-endpoint \
    --resource-group <resource-group>\
    --workspace-name <workspace-name> \
    --query "scoring_uri" \
    --output tsv

    Summary
    | Component | Value |
    | Workspace | <workspace-name> |
    | Endpoint Name | <endpoint-name> |
    | Deployment Name | <deployment-name> |
    | Region | <region>|
    | Scoring URL | https://<name>-endpoint.westus.inference.ml.azure.com/score
  15. Test the endpoint:
    az ml online-endpoint show --name helix-gpt-v7-25_3-endpoint
    az ml online-endpoint get-credentials --name helix-gpt-v7-25_3-endpoint
    To test:
    curl -X POST <scoring-uri> \
      -H "Authorization: Bearer <key>" \
      -H "Content-Type: application/json" \
      -d '{"input": "your input here"}'

For more information about deploying models on Microsoft Azure AI, see Microsoft Azure Machine Learning documentation

Task 3: To update model settings in BMC Helix Agent Studio

After deploying the model, copy the endpoint ID and the API key and perform the following steps:

  1. Log in to BMC Helix Innovation Studio.
  2. Select Workspace > HelixGPT Agent Studio.
  3. Select the Model record definition, and click Edit data. 
  4. On the Data editor (Model) page, search for the HelixGPT model name.
    The model ID is a unique identifier for the fine-tuned model. You are required to provide this exact ID to all agents in Task 5.  
  5. Select the model and click Edit.
  6. On the Edit record pane, turn off the Seed Data option, and provide the following information about the model that you deployed in your cloud environment:
    FieldDescriptionDefault or recommended value
    Auth TypeA unique authorization key used to ensure secure communication between BMC HelixGPT and the model.

    Google Cloud Platform Vertex AI: API Key

    Microsoft Azure AI: API Key

    Status

    Indicates the current operational state of the model. The following options are available:

    • New
    • Assigned
    • Fixed
    • Rejected 
    • Close
    		Select New.
    VendorThe name of the organization or provider offering the model.Supported providers: Google Cloud Platform, Vertex AI, or Microsoft Azure AI.

    API Endpoint Url

    		The specific URL to access the model.

    GCP Vertex AI: Endpoint ID

    Microsoft Azure AI: Azure ML Online Endpoint 

     

    API Key

    A unique key provided by the model vendor to authenticate API requests.

     

    GCP Vertex AI: Service account API key generated in Task 2.
    The service account must have the roles/aiplatform.user role. For more information, see Vertex AI roles and permissions.

    Microsoft Azure AI: Service account API key, encoded in a Base64 format.

    Default Config

    The predefined settings or parameters are applied to the model. 

    An administrator can modify the default configuration.

    GCP Vertex AI: Provide the information in the following format: 

    {
      "apiType": "vertexaimodelgarden",
      "deployedModelType": "HelixGPT-v6",
      "deploymentName": "<Google Cloud Project Name>",
      "location": "<Name of the Google Cloud region; example: us-central1>"
    }

    Microsoft Azure AI: Provide the information in the following format: 

    {{ "deploymentName": "helix-gpt-v7-25-3-deploy","deployedModelType":"HelixGPT",}}
    {{ "apiType": "azure_ml"}}

  7. Save changes.

Task 4: To enable BMC Helix ITSM fine-tuned model in TRA

  1. Log in to BMC Helix Innovation Studio.
  2. Select Workspace > HelixGPT Agent Studio.
  3. Select BMC Helix IT Service Management > Ticket Resolver Agent Skill.
  4. Select TRA_SUPERVISOR agent, and Clone the agent.
    TRA_AS.png
  5. Add a suitable suffix name for the new agent and select the Deep copy.
    In this case, the suffix used is Fine_Tune_Model_Support, so the fully qualified agent name is TRA_SUPERVISOR_Fine_Tune_Model_Support.
    Step1.1.jpg
  6. Select TRA_SUPERVISOR_Fine_Tune_Model_Support agent from the list, and click Apply.
    TRA_supervisor_agent.png
  7. In BMC Helix Innovation Studio click Agents > ITSM BAR Supervisor Agent_Fine_Tune_Model_Support
  8. Click Agent fleet and check box ITSM BAR FTM Agent.
    TRA_agent fleet.png
  9. Click Save.

Where to go from here

Using Ticket Resolver to manage incidents

 

Tip: For faster searching, add an asterisk to the end of your partial query. Example: cert*

BMC Helix ITSM 26.1