Optional ReadonlyinputAlarm when input tokens in one alarm period exceed this value. Omit to skip the input-token alarm.
Use cases: detecting oversized prompts or a surge in requests
AWS: CloudWatch alarm on InputTokenCount (Sum)
Validation: Optional; Number; >= 0
ReadonlymodelModel and inference-profile IDs whose token use the alarms sum, exactly as callers
pass them, e.g. amazon.nova-micro-v1:0 and us.amazon.nova-micro-v1:0. CloudWatch
matches the ModelId dimension exactly, so traffic sent under an ID not listed here
is not counted.
Use cases: covering the base model ID and each inference-profile ID callers use
AWS: AWS/Bedrock ModelId metric dimension
Validation: Required; String[]; 1-10 unique IDs
Optional ReadonlyoutputAlarm when output tokens in one alarm period exceed this value. Omit to skip the output-token alarm.
Use cases: detecting runaway generation, output-token spend spikes
AWS: CloudWatch alarm on OutputTokenCount (Sum)
Validation: Optional; Number; >= 0
Token-use alarms for one model, summed across the model IDs listed. Bedrock publishes token metrics per
ModelIdfor the whole account, keyed on the model or inference-profile ID exactly as the caller passed it, so the alarms count every caller in the account that uses a listed ID - and none that uses an unlisted one, such as another inference profile for the same model.Use cases: per-model token-spend alerting, covering a base model and its cross-Region inference profiles in one alarm
AWS: CloudWatch alarms on AWS/Bedrock InputTokenCount / OutputTokenCount by ModelId
Validation: modelIds required (1-10, unique); at least one threshold