Revert "Add metrics for speculative decoding (acceptance rate, average acceptance length)" (#11433)

This commit is contained in:
Scott Lee
2025-10-10 12:54:57 -07:00
committed by GitHub
parent b4408e6098
commit 55b14656e6
9 changed files with 1 additions and 74 deletions
-8
View File
@@ -127,7 +127,6 @@ class SchedulerStats:
# Speculative decoding
spec_accept_length: float = 0.0
spec_accept_rate: float = 0.0
# Retract
num_retracted_reqs: int = 0
@@ -221,12 +220,6 @@ class SchedulerMetricsCollector:
labelnames=labels.keys(),
multiprocess_mode="mostrecent",
)
self.spec_accept_rate = Gauge(
name="sglang:spec_accept_rate",
documentation="The average acceptance rate of speculative decoding (`accepted tokens / total draft tokens` in batch).",
labelnames=labels.keys(),
multiprocess_mode="mostrecent",
)
# Retract
self.num_retracted_reqs = Gauge(
@@ -527,7 +520,6 @@ class SchedulerMetricsCollector:
# Speculative decoding
self._log_gauge(self.spec_accept_length, stats.spec_accept_length)
self._log_gauge(self.spec_accept_rate, stats.spec_accept_rate)
# PD disaggregation
self._log_gauge(