Google Vertex AI Publisher Model metrics

A Vertex AI publisher model is a pre-built foundation model, such as a Google or Meta model, that you call through Vertex AI rather than deploy yourself. Each model version that your project invokes is monitored as a separate entity. Ensure your cloud platform is configured in SolarWinds Observability SaaS to collect this service's data. See Add a GCP cloud account.

Many of the collected metrics from Vertex AI Publisher Model entities are displayed as widgets in SolarWinds Observability explorers; additional metrics may be collected and available in the Metrics Explorer. You can also create an alert for when an entity's metric value moves out of a specific range. See Entities in SolarWinds Observability SaaS for information about entity types in SolarWinds Observability SaaS.

The following table lists some of the metrics collected for these entities. To see the Vertex AI Publisher Model metrics in the Metrics Explorer, type aiplatform.googleapis.com/publisher in the search box.

Metric	Unit	Description
sw.metrics.healthscore	Percent (%)	Health state. The health state provides real-time insight into the overall health and performance of your monitored entities. The health state is determined based on anomalies detected for the entity, alerts triggered for the entity's metrics, and the status of the entity. The health state is displayed as one of the following four states and colors: Good, Moderate, Bad, or Unknown. You can determine the impact of the alerts, anomalies, and statuses on the health of an entity type by going to Settings > Health, and selecting a specific entity type. You can also customize the impact. To view the health of Google Vertex AI Publisher Model entities in the Metrics Explorer, filter the sw.metrics.healthscore metric by entity_types and select gcpvertexaipublishermodel.
gcp.aiplatform.googleapis.com. publisher.onlineServing.modelInvocationCount	Count per second	Number of times the publisher model was invoked, per project.
gcp.aiplatform.googleapis.com. publisher.onlineServing.modelInvocationLatencies	Milliseconds (ms)	Latency distribution of publisher model invocations.
gcp.aiplatform.googleapis.com. publisher.onlineServing.firstTokenLatencies	Milliseconds (ms)	Time-to-first-token latency distribution for streaming model invocations.
gcp.aiplatform.googleapis.com. publisher.onlineServing.tokenCount	Count per second	Number of input and output tokens processed by the publisher model.
gcp.aiplatform.googleapis.com. publisher.onlineServing.tokens	Count	Input and output token count distribution for the publisher model.
gcp.aiplatform.googleapis.com. publisher.onlineServing.characterCount	Count per second	Accumulated input and output character count processed by the publisher model.
gcp.aiplatform.googleapis.com. publisher.onlineServing.characters	Count	Input and output character count distribution for the publisher model.
gcp.aiplatform.googleapis.com. publisher.onlineServing.consumedThroughput	Count per second	Character throughput consumed by the publisher model for the project.