inference/research
verifiedhigh · 0.95updated 2026-10-04

vLLM engine.arg_utils --disable-log-stats default (stats collection is on by default)

repo

Fields

id
sources/w3m-vllm-observe-latency-declaration
Publisher
vllm-project/vllm
Kind
repo
Accessed
2026-10-04
notes
vllm/engine/arg_utils.py on main, read 2026-10-04. EngineArgs.disable_log_stats: bool = False (line 605) with parser.add_argument("--disable-log-stats", action="store_true", help="Disable logging statistics.") at line 1863. The flag being a store_true over a False default is the primary-source basis for enabled_by_default=true on every vLLM metric_exposure record: statistics collection - and therefore the /metrics population - is on unless someone explicitly turns it off. Paired with this, vllm/config/observability.py sets kv_cache_metrics: bool = False with the docstring 'Enable KV cache residency metrics (lifetime, idle time, reuse gaps). Uses sampling to minimize overhead. Requires log stats to be enabled (i.e., --disable-log-stats not set).' and kv_cache_metrics_sample default 0.01.

Notes

vllm/engine/arg_utils.py on main, read 2026-10-04. EngineArgs.disable_log_stats: bool = False (line 605) with parser.add_argument("--disable-log-stats", action="store_true", help="Disable logging statistics.") at line 1863. The flag being a store_true over a False default is the primary-source basis for enabled_by_default=true on every vLLM metric_exposure record: statistics collection - and therefore the /metrics population - is on unless someone explicitly turns it off. Paired with this, vllm/config/observability.py sets kv_cache_metrics: bool = False with the docstring 'Enable KV cache residency metrics (lifetime, idle time, reuse gaps). Uses sampling to minimize overhead. Requires log stats to be enabled (i.e., --disable-log-stats not set).' and kv_cache_metrics_sample default 0.01.

Cross-references

This record references no other records.