minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Tiny add explanations to realtime token metric labels (#16009)

Browse Source
This commit is contained in:
fzyzcjy
2026-01-03 12:36:55 +08:00
committed by GitHub
parent 078d96213a
commit 9821fae5c6
1 changed files with 16 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/metrics/collector.py
+16 -4
View File
@@ -729,23 +729,35 @@ class SchedulerMetricsCollector:
self.realtime_tokens_total = Counter( self.realtime_tokens_total = Counter(
name="sglang:realtime_tokens_total", name="sglang:realtime_tokens_total",
documentation="Total number of tokens processed (updated on each log interval).", documentation=(
"Total number of tokens processed (updated on each log interval). "
"mode: prefill_compute, prefill_cache, decode."
),
labelnames=list(labels.keys()) + ["mode"], labelnames=list(labels.keys()) + ["mode"],
) )
self.gpu_execution_seconds_total = Counter( self.gpu_execution_seconds_total = Counter(
name="sglang:gpu_execution_seconds_total", name="sglang:gpu_execution_seconds_total",
documentation="Total time that GPU is busy executing a workload.", documentation=(
"Total time that GPU is busy executing a workload. "
"Refer to ForwardMode for category labels."
),
labelnames=list(labels.keys()) + ["category"], labelnames=list(labels.keys()) + ["category"],
) )
self.dp_cooperation_realtime_tokens_total = Counter( self.dp_cooperation_realtime_tokens_total = Counter(
name="sglang:dp_cooperation_realtime_tokens_total", name="sglang:dp_cooperation_realtime_tokens_total",
documentation="Total number of tokens processed with labels about DP cooperation.", documentation=(
"Total number of tokens processed with labels about DP cooperation. "
"mode: prefill_compute, prefill_cache, decode."
),
labelnames=list(labels.keys()) + ["mode", "num_prefill_ranks"], labelnames=list(labels.keys()) + ["mode", "num_prefill_ranks"],
) )
self.dp_cooperation_gpu_execution_seconds_total = Counter( self.dp_cooperation_gpu_execution_seconds_total = Counter(
name="sglang:dp_cooperation_gpu_execution_seconds_total", name="sglang:dp_cooperation_gpu_execution_seconds_total",
documentation="Total time that GPU is busy executing a workload with labels about DP cooperation.", documentation=(
"Total time that GPU is busy executing a workload with labels about DP cooperation. "
"Refer to ForwardMode for category labels."
),
labelnames=list(labels.keys()) + ["category", "num_prefill_ranks"], labelnames=list(labels.keys()) + ["category", "num_prefill_ranks"],
) )