[Vertex AI] Implement Two Tower Model training with a custom job

Two Tower Model learning is implemented using a custom job.
However, the model artifact is not output to GCS even after the custom job completes.
I checked the bucket permissions and other custom job request details, but there is no improvement.
Is there anyone experiencing a similar event?

It sounds like you’re encountering issues with a custom job not outputting the model artifact to Google Cloud Storage (GCS) even after completion. Troubleshooting such issues can involve checking several aspects:

Job Configuration: Ensure that the output path specified in the custom job configuration is correct and accessible. Any discrepancies here could result in the artifact not being stored in GCS.

Permissions: Verify the permissions on the GCS bucket. The service account or credentials used by the job might lack the necessary permissions to write to the specified GCS location. Double-check and ensure that the service account has the required access rights.

Logging & Error Handling: Review the job logs or any error messages generated during or after the job execution. This could provide specific clues about why the artifact isn’t being stored in GCS. Look for any permission-related errors or issues with the output path.

Networking & Connectivity: Sometimes, networking issues might prevent successful uploads to GCS. Check if there are any network restrictions or firewalls that could be hindering the communication between your job execution environment and GCS.

If none of these steps seem to resolve the issue, reaching out to the support channels of the platform or service you’re using for custom jobs could be the next step. Sharing specific error messages or logs could help them diagnose and assist in resolving the problem more effectively.

Poala_Tenorio, thank you for your reply.

I checked the troubleshooting you pointed out, but no model artifact was output to GCS.

Regarding the GCS output path, even if I set a path that does not exist, it completed normally without any error.

I would appreciate it if you could give me some advice if you have any ideas from this point of view.

thank you.