Application Lifecycle Management : AI Builder Model and Training Data Set missing in Azure Repository Source Control

[Rajesh365]

Hi All, 

I am working on setting up an Application Lifecycle Management where I have created a solution within which I have a Power Automate and an AI Builder Custom form processing Model which is consumed within the flow. 

I have managed to successfully move the Power Automate and AI Builder model from Dev to Prod environment using Azure DevOps Pipelines and in between, I am using the Azure Repos to Source Control the solution contents. 

However, neither the AI Builder Model or its Training Data set is visible in the Source Control. I can only see the Power Automate data. Once Build/Release pipelines runs, the AI builder is still successfully migrated to Prod and it works as expected within the Power Automate in the target.

Is there something I am missing to do , due to which the AI Builder Model and Its training data set is not present in the Azure Repos Source Control ? 

Dev Unmanaged Solution : 

Azure Repo Source Control : 

Thanks in Advance !

[JoeF-MSFT]

Hi @Rajesh365 - nice ALM setup, thanks for sharing!

What you're seeing is correct. When you export an AI Builder model, you're actually exporting the metadata around the model, but not the model binary itself, neither the training data. When you import the solution into another environment, we use that metadata from the model to make a copy of the underlying model hosted in Azure so to make it available in the imported environment. The training data is not copied, which means that you can use the model in the environment where it has been imported but not edit the model, as the training data won't be there. 

The recommendation is to update the model on the original environment, and then promote the model to other environments. With your setup, you can automate it very nicely. 

Hope this helps clarify. 

[Rajesh365]

Thanks a lot @JoeF-MSFT  for the quick response and helping us understand what happens behind the scenes. 

Just a follow up question :

The Share Option, as per my limited understanding, lets other users to use a model I have created in their Apps and flows but does not let them edit the model and train them better. In case we want to provide multiple users the ability to edit and train a model in the Dev environment, what are the current possibilities ?

Thanks in Advance !

[JoeF-MSFT]

You're welcome @Rajesh365 !Your understanding is correct around sharing. The good news is that we're working on enabling sharing form processing models with multiple owners that can edit. We plan to release this by October.  

[Rajesh365]

That's Awesome news @JoeF-MSFT . Sorry to trouble with a few more follow up questions. 

• From the official docs, I did see that we can assign System Customizer role in case we want to share the model with others who need to edit and train the model. But in a business scenario where we may have 10+ business users whom we are trying to empower with System Customizer role so that they can edit some of the commonly used AI models, does that sound like a best practice as they may end up getting un intended access to else where in the environment ? 
• In case we want to move the training data and AI Model from an old Dev environment to a newly provisioned Dev environment to continue the development in the new target, do we have any easy mechanism to do it as Azure Pipelines do not support this functionality as of now ? Or is there a way to move some of the AI Model specific tables to the new environment to migrate the models along with training content (just a wild thought)?

Again, Thanks a lot for your help and insights 🙂

[JoeF-MSFT · Accepted solution]

Hi @Rajesh365 - your questions are great feedback! 😊

• Do you remember where you saw the recommendation around using System Customizer as a way to have multiple owners of the model? We don't intend to promote it as a best practice. What we provide is a way to transfer model ownership like this: https://docs.microsoft.com/en-us/ai-builder/share-model#the-owner-of-a-model-has-left-the-company-ho...
• Unfortunately, this is not supported yet. We keep hearing this ask from customers, so we're thinking in ways to enable this in a future update. 

You're welcome and additional questions are more than welcome. On the AI Builder team, we're hosting an AMA event next week with various experts from the team: Announcing the June 7th AI Builder AMA - Power Platform Community (microsoft.com) We're happy to take more ALM questions and discuss live there 🙂

[Rajesh365]

You are right @JoeF-MSFT  . I saw the same link that you had shared where we can use System Customizer to transfer ownership, guess I mistook that as an opportunity for multiple people to edit the model. 

So happy to have interacted with you and looking forward to the AMA. Thank You !