All versions · Version range: v0.23.0

vllm:prompt_tokens_cached

Counter for cached prompt tokens, local plus external.

MetricPrometheusverified Sep 29

Version history

The only version covered here is v0.23.0, where vllm:prompt_tokens_cached is registered as a Prometheus counter with the label names model_name and engine under the PrometheusStatLogger V1 metrics publisher.
In v0.23.0 it counts cached prompt tokens reused during processing, combining local and external sources.
In v0.23.0 the counter increments from the per-iteration PromptTokenStats.cached_tokens field.
In v0.23.0 it represents aggregate saved prompt-token work and is not by itself a per-request cache-hit percentage.
For v0.23.0, the guidance is to compare its rate with prompt-token demand under matching labels and windows, separating local and external attribution when required.

Identity

Prometheus type: counter
Labels: model_nameengine
Unit: Not recorded — vLLM registration code has no structured unit field, and units are not inferred from name suffixes

Semantic coverage

covered · Availability & namecovered · Structural identitycovered · Measurement semanticscovered · Causal/runtime semanticscovered · Interpretation & compositioncovered · Guarded guidance

Metric knowledge (6)

Claim	Semantic layer	Verified
In vLLM v0.23.0, `vllm:prompt_tokens_cached` counts cached prompt tokens reused during processing, combining local and external sources.	Measurement semantics	Sep 29
In vLLM v0.23.0, `vllm:prompt_tokens_cached` increments from the iteration `PromptTokenStats.cached_tokens` field.	Causal/runtime semantics	Sep 29
In vLLM v0.23.0, `vllm:prompt_tokens_cached` is aggregate saved prompt-token work and is not by itself a per-request cache-hit percentage.	Interpretation & composition	Sep 29
In vLLM v0.23.0, `vllm:prompt_tokens_cached` compare its rate with prompt-token demand under matching labels and windows; separate local/external attribution when required.	Guarded guidance	Sep 29
In vLLM v0.23.0, `vllm:prompt_tokens_cached` is registered as a Prometheus counter.	Structural identity	Sep 29
In vLLM v0.23.0, `vllm:prompt_tokens_cached` is registered with the label names `model_name`, engine.	Structural identity	Sep 29

Related Use Cases (1)

#40696: Full-block eligibility creates prefix-cache cliffs

Availability

vllm:prompt_tokens_cached is registered as a Prometheus counter in vLLM v0.23.0.

vllm.surface.prometheus_metricscomplete_inventoryPrometheusStatLogger metrics publisher

vLLM v0.23.0 v1 metrics loggers.py · vllm/v1/metrics/loggers.py:653-660