A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 13:30 UTC
[Metrics] Fix minor inconsistencies in bucket progression (#17262)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Apr 28, 2025 at 00:19 UTC
4213475ec7b05a0287d7ec044d1c7afa2493ba28
2 files changed
+3
-3
vllm/engine/metrics.py
+1
-1
@@ -145,7 +145,7 @@ class Metrics:
145
documentation="Histogram of number of tokens per engine_step.",
146
labelnames=labelnames,
147
buckets=[
148
- 1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8096, 16192
148
+ 1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384
149
])
150
self.histogram_time_to_first_token = self._histogram_cls(
151
name="vllm:time_to_first_token_seconds",
vllm/v1/metrics/loggers.py
+2
-2
@@ -233,8 +233,8 @@ class PrometheusStatLogger(StatLoggerBase):
233
name="vllm:iteration_tokens_total",
234
documentation="Histogram of number of tokens per engine_step.",
235
buckets=[
236
- 1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8096,
237
- 16192
236
+ 1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192,
237
+ 16384
238
],
239
labelnames=labelnames).labels(*labelvalues)
240