Unable to Publish NLU model in UAT

Yasmin · ‎08-09-2021

When migrating VA and NLU updates into UAT environment via updateset we are able to train the model, but it does not allow to publish the model. With the migration, there were some new utterances included, so we need to train and publish, but it only allows to train and the publish button is not enabled. What could be causing this? Thanks.

Yasmin · ‎08-11-2021

Thanks brianbakker. We tried eliminating the 0% progress record from the ML_solution.list but it still wouldn't allow to publish after training it.

Somehow this was related to another NLU issue that I reported on this thread "VA topic not mapped to NLU Intent" where the name of one the topics was the same name as the intent. Though this wasn't an issue before, it now seems to not let NLU mapped objects to work properly. Once I changed the name of the intent to be different from the VA topic name, I was able to train and publish the the model successfully.

Link to related article that solved this issue: VA topic not mapped to NLU Intent

View solution in original post

mpmacconnell · ‎08-09-2021

I have a similar problem with the out of the box ITSM NLU for Virtual Agent NLU. But in my case it just says "Cannot convert null to an object." and will not let me Publish. I have had a HI ticket open for a very long time on this and they can't seem to figure it out. Nothing changed to this one except I did a recent update of the plugin. We have not modified the out of the box version, but we do use the intents in some of the topics we cloned to modify for our enviroment.

Chris D · ‎11-03-2021

Did you ever get a resolution? Just noticed this "Cannot convert null to an object." issue with training the ootb "[Read Only] Setup Topics Model" in one of our instances - though it trained and published just fine in two other instances.

Brian Bakker · ‎11-04-2021

Hello Chris,

This error is usually caused because of the missing associated attachments for the [ml_model_artifact] records are not preserved on the affected instance after the clone.

To determine which [ml_model_artifact] records are associated with the NLU Model, use table [sys_nlu_model] for the ML Solution name. Use table [ml_solution] to find the latest version of the ML Solution name. Using the sys_id of the [ml_solution] record, you can then find the associated [ml_model_artifact] records related to the trained solution. If any of these [ml_model_artifact] records are missing the associated [sys_attachment] record, then there will be an issue with this NLU Model.

You can export these records from the source instance that have the sys_attachment record and import them into the instance which are missing these attachments.

If the issue persists, please raise a case with ServiceNow Technical Support.

Best regards,

Brian

Brian Bakker · ‎08-11-2021

Please review KB0952640: NLU Model training throws an error without an error message and creates a record in table [ml_solution] for the trained NLU Model solution, which is inactive with 0% progress, but the state is set to "Solution Complete" causing further training issues.

You need to be logged in to support.servicenow.com to view this KB article.