You’re offline. This is a read only version of the page.
77
Hi,
I've moved a model to another environment, by placing it in a solution, exporting that solution (unmanaged) and importing it into the target environment.
I didn't get any errors during import, but now, after importing, I cannot edit the model anymore. The button is grayed out. And it cannot display documents...
Did I forget to include anything in the solution?
Please advise...
Best regards,
Ernst Wolthaus
Hi Antonie,
I do have an issue with AI Builder. Please help me out to solve it.
I have created AI builder in a Dev environment. But after a while got to know that was not the correct Dev environment. Now if I move existing AI builder to new environment then we are not able to edit it for re-training purpose. Every time I need to go back to old environment, re-train model and export/import whole solution to new dev environment.
It will be a great help if now we have the solution to edit AI builder once we import it to other environment.
Hi Antoine,
I have managed to copy below tables and also maintaining lookups from UAT to PROD. Still I am not able to edit/ create new versions of the model in Prod.
1. Dataset Container
2. Dataset
3. File
4. Dataset File
5. File Attached Data
Am I missing any table?
Thanks,
Pankaj
Unfortunately we do not have an easy to document means of moving training data across environments at the moment. If your dataset is small enough, I would recommend just starting from scratch in the new environment. If it's too big to start from scratch, we can try to provide a description of tables & rows which would need to be copied but it's fairly complex to get right.
Regarding your second scenario, your statement is correct. Training data is only required to train the model. When you import the model in a new environment you can use it in an app and a flow even if the training data is not available.
The whole topic of Model Operations which your questions relate to is something we've started formalizing better from a vision and implementation sequencing perspective. I hope that we'll be able to deliver early improvements over the first half of 2022 to support such cross environment data movement easily. At a gist, if you're working with development, testing and production environments, you want to be able to move models up from development to testing and then production, which solutions allow today. You also want to be able to move training data down from production to development, for which we don't have an easy to use mechanism today.
Ah, right, I understand now...
So, I needed to move my model to another environment because we decided the current environment wasn't suitable for a PoC I'm working on the AI Builder form processing.
So I want to move the model which I have already trained to the new environment and train it a bit more.
And even if I didn't need to add any more collections or training documents, it would be important to us to look at the 'training material' in the model.
In another scenario: we would create and train a model and then move it to Production. In that case, we would only need the model on Production, apps and flows would be able to use it, without the underlying data... right?
And finally, we might add documents/collections/fields to the model and train the model in Production, but changing the model 'on the fly' in Production would not be the preferred scenario.
So, what would you advise?
Best regards,
Ernst Wolthaus
Hi Ernst,
It is unfortunately not possible to create a new version of a Forms Processing model after importing it as the import process only copies the model, not the underlying training data. As we're currently looking at making training data copying possible easily, I'd be curious to understand your workflow so we can account for it in future product updates.
More information on such limitations can be found at https://docs.microsoft.com/en-us/ai-builder/distribute-model#limitations .
Antoine
Share
Report
All responses (
Answers (
