The Generative AI feature in Business Automation Workflow 24.0.1 supports foundation models provided with watsonx.ai in IBM watsonx.ai as a Service. This allows customers to start exploring integrating Generative AI into their workflows.  However, these foundation models are hosted in a multitenant hardware environment and follow the watsonx.ai foundation model lifecycle, which includes a deprecation policy. Consequently, these models often have short lifecycles and low inference rate limits, which may not align with the demands of a production environment in a conventional workflow context.

With Business Automation Workflow 25.0.0, these constraints have been eliminated because the Generative AI feature adds support for custom foundation models and deploy on demand foundation models. Furthermore, for those customers who like to deploy and maintain their foundation models on-premises, it adds support for custom foundation models, alongside the provided foundation models in IBM watsonx.ai Software in IBM Cloud Pak for Data.

Enabling these supports in Business Automation Workflow 25.0.0 requires almost the same steps as described in the Webinar Using Generative AI and Assistant in workflows for Business Automation Workflow 24.0.1.  The subsequent sections provide additional details on the process of enabling the Generative AI feature to support custom foundation models and deploy on demand models.

Enabling custom foundation models, deploy on demand foundation models, and the provided foundation models in IBM watsonx.ai as a Service

To enable the custom foundation models, the deploy on demand foundation models, and the provided foundation models in IBM watsonx.ai as a Service in the Generative AI feature, for your authoring and production environments, you must:

1. Deploy the custom foundation model and the deploy on demand foundation model in your watsonx.ai account.
   • Ensure to include a serving name in your deployment
2. Set up the authentication alias as described in the webinar
3. When updating the 100Custom.xml, in addition to providing your account's project ID, provider URL, and credential information, you must also include:
   • The types of foundation models you need to include by using the supported-foundation-model-type element
   • If you are using deploy on demand foundation models, you must also include the space-id element because these foundation models are deployed in your deployment space.
   • For a detailed description of each element, refer to Additional content to learn how to enable the Generative AI feature in detail

Below is an example of the 100Custom.xml snippet for enabling custom foundation models, deploy on demand foundation models, and the provided foundation models in IBM watsonx.ai as a Service.

<properties>
  <server>
    <gen-ai merge="mergeChildren">
            <provider-url>https://us-south.ml.cloud.ibm.com</provider-url>
            <project-id>00000000-1111-2222-3333-444444444444</project-id>
            <auth-alias>watsonAI.alias</auth-alias>
            <space-id>12345678-1111-2222-3333-444444444444</space-id>
            <supported-foundation-model-type>watsonx</supported-foundation-model-type>
            <supported-foundation-model-type>custom</supported-foundation-model-type>
            <supported-foundation-model-type>curated</supported-foundation-model-type>
    </gen-ai>
  </server>
</properties>

Once your changes are synchronized and the workflow server is started, your available foundation models and their corresponding deployment types are displayed in your Service Flow Gen AI tab in the Process Designer.  Below is a screen capture that shows a view where all three types of foundation models are available to use in the Generative AI task.

The tooltip also displays the types of foundation models that are available for you to use in the Large language model list.

Enabling custom foundation models, and the provided foundation models in IBM watsonx.ai Software in IBM Cloud Pak for Data

To enable the custom foundation models and the provided foundation models in IBM watsonx.ai Software in IBM Cloud Pak for Data, for your authoring and production environments, you must:

1. Deploy the custom foundation model in your IBM Cloud Pak for Data environment
   • Ensure to include a serving name in your deployment
2. Set up the authentication alias as described in the webinar
3. When updating the 100Custom.xml, in addition to providing your account's project ID, provider URL, and credential information in the 100Custom.xml, you must also include:
   • The authentication URL to your CP4D environment by using the auth-url element
   • The authentication type by using the auth-type element with the CP4D value
   • (Optional) If the server certificate from the provider-url and the auth-url are not signed by any of the public certification authorities included in the preconfigured SSL configuration, an administrator must create a new SSL configuration for this service, import the server certificate into a new trust store, and update the new SSL configuration name using the ssl-configuration element.
   • The types of foundation models you need to include by using the supported-foundation-model-type element.
   • For a detailed description of each element, refer to Additional content to learn how to enable the Generative AI feature in detail

Below is an example of the 100Custom.xml snippet for enabling custom foundation models, and the provided foundation models in IBM watsonx.ai Software in IBM Cloud Pak for Data.

<properties>
    <server>
        <gen-ai merge="mergeChildren">
            <project-id>00000000-1111-2222-3333-444444444444</project-id>
            <provider-url>https://my-cpd.mycompany.com</provider-url>
            <auth-alias>watsonx.ai_auth_alias</auth-alias>
            <auth-url merge="replace">https://my-cpd.mycompany.com/icp4d-api/v1/authorize</auth-url>
            <auth-type merge="replace">CP4D</auth-type>
            <ssl-configuration merge="replace">MyNewSSLSettings</ssl-configuration>
            <supported-foundation-model-type>watsonx</supported-foundation-model-type>
            <supported-foundation-model-type>custom</supported-foundation-model-type>
        </gen-ai>
    </server>
</properties>

Your available foundation models and their corresponding deployment types are displayed in your Service Flow Gen AI tab in the Process Designer.  Below is a screen capture that shows a view of the two types of foundation models that are available to use in the Generative AI task.

Conclusion

By enabling the use of custom or deploy on demand foundation models within a Generative AI task, organizations gain flexibility and long-term viability. This approach ensures your tasks remain resilient since your models remain available as long as your foundation model deployments are available. Plus, it offers a significant performance advantage over the provided foundation models in IBM watsonx.ai as a Service due to their dedicated deployments without rate limits, allowing for greater scalability and responsiveness. Furthermore, organizations can set up and deploy their foundation models within their IBM Cloud Pak for Data on-premises environment, ensuring data remains securely on-premise.

Additional content

• To learn how to enable the Generative AI feature in detail:
  • see Webinar Using Generative AI and Assistant in workflows
  • see Enabling generative AI in Business Automation Workflow (traditional)
  • see Enabling generative AI in Business Automation Workflow (containers)
• For feature differences between IBM watsonx.ai as a Service and IBM watsonx.ai Software, see here
• For the different foundation model deployment types in IBM watsonx.ai, see here