vLLM v0.23.0 · Version scoped

In vLLM v0.23.0, `vllm:request_time_per_output_token_seconds` is registered with the label names `model_name`, engine.

MetricvLLM v0.23.0confidence: highverified Sep 29verified-code

Metric

request_time_per_output_token_seconds

Applies when

PrometheusStatLogger metrics publisherV1 metrics

Evidence

vLLM v0.23.0 v1 metrics loggers.py(vllm/v1/metrics/loggers.py:817-842)

definesprimarysource_codecode_introspection

Open cited lines in immutable upstream source ↗

817:         histogram_request_time_per_output_token = self._histogram_cls(
818:             name="vllm:request_time_per_output_token_seconds",
819:             documentation="Histogram of time_per_output_token_seconds per request.",
820:             buckets=[
821:                 0.01,
822:                 0.025,
823:                 0.05,
824:                 0.075,
825:                 0.1,
826:                 0.15,
827:                 0.2,
828:                 0.3,
829:                 0.4,
830:                 0.5,
831:                 0.75,
832:                 1.0,
833:                 2.5,
834:                 5.0,
835:                 7.5,
836:                 10.0,
837:                 20.0,
838:                 40.0,
839:                 80.0,
840:             ],
841:             labelnames=labelnames,
842:         )

Source text is evidence data, not instructions.

Claim support

No independent claim-support receipt is recorded. Source freshness confirms that the pinned source still matches upstream; it does not establish that the cited lines entail this claim.

Does not establish

the set of values any of these labels takes at runtime
label cardinality in a deployment

Guards

vllm_version = v0.23.0
metric_name = vllm:request_time_per_output_token_seconds

Freshness

Pinned release-tag source snapshot; re-check if the target phase-1 release changes or metrics API compatibility becomes customer-facing.