Delen via

GenerationTokenStatisticsSignal Class

Note

This is an experimental class, and may change at any time. Please see https://aka.ms/azuremlexperimental for more information.

Generation token statistics signal definition.

Constructor

GenerationTokenStatisticsSignal(*, production_data: LlmData | None = None, metric_thresholds: GenerationTokenStatisticsMonitorMetricThreshold | None = None, alert_enabled: bool = False, properties: Dict[str, str] | None = None, sampling_rate: float | None = None)

Keyword-Only Parameters

Name Description
production_data

input dataset for monitoring.

Default value: None
metric_thresholds
Optional[GenerationTokenStatisticsMonitorMetricThreshold]

Metrics to calculate and their associated thresholds. Defaults to App Traces

Default value: None
alert_enabled
bool

Whether or not to enable alerts for the signal. Defaults to False.

Default value: False
properties
Optional[Dict[str, str]]

The properties of the signal

Default value: None
sampling_rate
float

The sample rate of the target data, should be greater than 0 and at most 1.

Default value: None

Examples

Set Token Statistics Monitor.


   spark_compute = ServerlessSparkCompute(instance_type="standard_e4s_v3", runtime_version="3.4")
   monitoring_target = MonitoringTarget(
       ml_task=MonitorTargetTasks.QUESTION_ANSWERING,
       endpoint_deployment_id=f"azureml:{endpoint_name}:{deployment_name}",
   )
   monitoring_target = MonitoringTarget(
       ml_task=MonitorTargetTasks.QUESTION_ANSWERING,
       endpoint_deployment_id=f"azureml:{endpoint_name}:{deployment_name}",
   )
   monitor_settings = MonitorDefinition(compute=spark_compute, monitoring_target=monitoring_target)
   model_monitor = MonitorSchedule(
       name="qa_model_monitor", trigger=CronTrigger(expression="15 10 * * *"), create_monitor=monitor_settings
   )
   ml_client.schedules.begin_create_or_update(model_monitor)

Variables

Name Description
type
str

The type of the signal. Set to "generationtokenstatisticssignal" for this class.