inference/research
verifiedhigh · 0.93updated 2026-10-04

SGLang: a KV-cache usage gauge under vLLM's name - VERIFIED ABSENT

sglang · unknown

Fields

id
metric-exposures/w3m-sglang-no-sglang-kv-cache-usage-perc
Metric name
sglang:kv_cache_usage_perc
Metric type
unknown
Exposure
unknown
Verified negatively
yes
notes
VERIFIED NEGATIVE. SGLang does not emit a KV-cache usage gauge under vLLM's name. sglang:kv_cache_usage_perc is NOT a metric this engine registers - it is recorded here under the conventional name a reader would go looking for so the absence is queryable rather than implied by a gap in a list. SGLang exposes KV occupancy as a pool DECOMPOSITION instead - sglang:token_usage (max across pools, flagged misleadingly named in the source itself) plus full/swa/mamba_usage and the available/evictable/used absolute counts. Strictly more information, under different names. What was searched: grep for name="sglang:kv_cache_usage_perc" returns zero hits.. Do not read this record's metric_name as something the engine exports; it is the name that is missing.

Notes

VERIFIED NEGATIVE. SGLang does not emit a KV-cache usage gauge under vLLM's name. sglang:kv_cache_usage_perc is NOT a metric this engine registers - it is recorded here under the conventional name a reader would go looking for so the absence is queryable rather than implied by a gap in a list. SGLang exposes KV occupancy as a pool DECOMPOSITION instead - sglang:token_usage (max across pools, flagged misleadingly named in the source itself) plus full/swa/mamba_usage and the available/evictable/used absolute counts. Strictly more information, under different names. What was searched: grep for name="sglang:kv_cache_usage_perc" returns zero hits.. Do not read this record's metric_name as something the engine exports; it is the name that is missing.

Cross-references