vLLM v0.23.0 · Version scoped

In vLLM v0.23.0, `vllm:request_prefill_kv_computed_tokens` is computed in finished-request logging after request completion from `num_prompt_tokens` and `num_cached_tokens`.

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

Metric

request_prefill_kv_computed_tokens

Applies when

PrometheusStatLoggerV1 metrics

Evidence

vLLM v0.23.0 v1 metrics loggers.py(vllm/v1/metrics/loggers.py:1179-1213)

definesprimarysource_codecode_introspection

Open cited lines in immutable upstream source ↗

1179:         for finished_request in iteration_stats.finished_requests:
1180:             self.counter_request_success[finished_request.finish_reason][
1181:                 engine_idx
1182:             ].inc()
1183:             self.histogram_e2e_time_request[engine_idx].observe(
1184:                 finished_request.e2e_latency
1185:             )
1186:             self.histogram_queue_time_request[engine_idx].observe(
1187:                 finished_request.queued_time
1188:             )
1189:             self.histogram_prefill_time_request[engine_idx].observe(
1190:                 finished_request.prefill_time
1191:             )
1192:             self.histogram_inference_time_request[engine_idx].observe(
1193:                 finished_request.inference_time
1194:             )
1195:             self.histogram_decode_time_request[engine_idx].observe(
1196:                 finished_request.decode_time
1197:             )
1198:             # Calculate prefill KV compute (excludes cached tokens)
1199:             prefill_kv_computed = finished_request.num_prompt_tokens - max(
1200:                 finished_request.num_cached_tokens, 0
1201:             )
1202:             self.histogram_prefill_kv_computed_request[engine_idx].observe(
1203:                 prefill_kv_computed
1204:             )
1205:             self.histogram_num_prompt_tokens_request[engine_idx].observe(
1206:                 finished_request.num_prompt_tokens
1207:             )
1208:             self.histogram_num_generation_tokens_request[engine_idx].observe(
1209:                 finished_request.num_generation_tokens
1210:             )
1211:             self.histogram_request_time_per_output_token[engine_idx].observe(
1212:                 finished_request.mean_time_per_output_token
1213:             )

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

a workload-independent healthy threshold
semantics in any vLLM version other than v0.23.0

Guards

vllm_version = v0.23.0
metric_name = vllm:request_prefill_kv_computed_tokens

Freshness

Pinned release-tag semantics; do not carry to another version without continuity evidence.