Closed devinmnorris closed 6 months ago
Thanks for opening your first issue here! Be sure to follow the issue template! If you are willing to raise PR to address this issue please do so, no need to wait for approval.
cc @MaksYermak can you take a look? I think you completed the system tests for Vertex AI. if the tests passes then maybe we are missing some coverage with this bug?
Hi @devinmnorris !
Regarding your example from section "Anything else", can you please provide the value from PARENT_MODEL_ID
parameter?
As far as i see from the implementation we have, the operator indeed takes only model_id
as input parameter, not the resource_name
Hi @VladaZakharova :)
We tried the SDK and the Operator approach using both:
model_id
i.e., 1234567890
resource_name
i.e., projects/1234/locations/us-central1/models/1234567890
It seems that either work when using the SDK, and neither work when using the Operator.
Apache Airflow Provider(s)
google
Versions of Apache Airflow Providers
apache-airflow-providers-google==10.12.0
Apache Airflow version
2.6.3
Operating System
Ubuntu 22.04.3 LTS
Deployment
Docker-Compose
Deployment details
No response
What happened
When creating AutoML Text Training jobs using
CreateAutoMLTextTrainingJobOperator
and providing the resource name or model ID of an existing model to theparent_model
parameter, an entirely new model withVersion 1
shows up in Vertex AI Model Registry.What you think should happen instead
Since we provided an argument to
parent_model
, the model uploaded by the job should be a version of the existing parent model.How to reproduce
If your model registry already has an existing model to use as the parent model, skip to step 3. Otherwise:
parent_model=initial_model_resource_name
Anything else
This problem only occurs when using the
CreateAutoMLTextTrainingJobOperator
, and not with the Vertex AI SDK for Python. For example, we were able to implement model versioning successfully using something like:google-cloud-aiplatform==1.41.0
Are you willing to submit PR?
Code of Conduct