xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-24 18:25:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V0][Metrics] Deprecate some KV/prefix cache metrics (#14136)

Browse Source 
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
This commit is contained in:
Mark McLoughlin 2025-03-03 18:25:46 +00:00 committed by GitHub
parent c41d27156b
commit 2dfdfed8a0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 25 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
vllm/engine/metrics.py
View File

@ -74,31 +74,51 @@ class Metrics:
],
multiprocess_mode="livemostrecent",
)
# Deprecated in 0.8 - KV cache offloading is not used in V1
# TODO: in 0.9, only enable if show_hidden_metrics=True
self.gauge_scheduler_swapped = self._gauge_cls(
name="vllm:num_requests_swapped",
documentation="Number of requests swapped to CPU.",
documentation=(
"Number of requests swapped to CPU. "
"DEPRECATED: KV cache offloading is not used in V1"),
labelnames=labelnames,
multiprocess_mode="sum")
# KV Cache Usage in %
self.gauge_gpu_cache_usage = self._gauge_cls(
name="vllm:gpu_cache_usage_perc",
documentation="GPU KV-cache usage. 1 means 100 percent usage.",
labelnames=labelnames,
multiprocess_mode="sum")
# Deprecated in 0.8 - KV cache offloading is not used in V1
# TODO: in 0.9, only enable if show_hidden_metrics=True
self.gauge_cpu_cache_usage = self._gauge_cls(
name="vllm:cpu_cache_usage_perc",
documentation="CPU KV-cache usage. 1 means 100 percent usage.",
documentation=(
"CPU KV-cache usage. 1 means 100 percent usage. "
"DEPRECATED: KV cache offloading is not used in V1"),
labelnames=labelnames,
multiprocess_mode="sum")
# Prefix caching block hit rate
# Deprecated in 0.8 - KV cache offloading is not used in V1
# TODO: in 0.9, only enable if show_hidden_metrics=True
self.gauge_cpu_prefix_cache_hit_rate = self._gauge_cls(
name="vllm:cpu_prefix_cache_hit_rate",
documentation="CPU prefix cache block hit rate.",
documentation=(
"CPU prefix cache block hit rate. "
"DEPRECATED: KV cache offloading is not used in V1"),
labelnames=labelnames,
multiprocess_mode="sum")
# Deprecated in 0.8 - replaced by queries+hits counters in V1
# TODO: in 0.9, only enable if show_hidden_metrics=True
self.gauge_gpu_prefix_cache_hit_rate = self._gauge_cls(
name="vllm:gpu_prefix_cache_hit_rate",
documentation="GPU prefix cache block hit rate.",
documentation=("GPU prefix cache block hit rate. "
"DEPRECATED: use vllm:gpu_prefix_cache_queries and "
"vllm:gpu_prefix_cache_queries in V1"),
labelnames=labelnames,
multiprocess_mode="sum")