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

Stop returning the unused prefix-computed flag from priority calc (#25718)

Browse Source
This commit is contained in:
fzyzcjy
2026-05-19 09:20:00 +08:00
committed by GitHub
parent 2d868656d0
commit 1cba3ab467
1 changed files with 2 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/schedule_policy.py
+2 -5
View File
@@ -160,19 +160,17 @@ class SchedulePolicy:
def calc_priority( def calc_priority(
self, waiting_queue: List[Req], running_batch: Optional[ScheduleBatch] = None self, waiting_queue: List[Req], running_batch: Optional[ScheduleBatch] = None
) -> bool: ) -> None:
if self.policy == CacheAgnosticPolicy.FCFS: if self.policy == CacheAgnosticPolicy.FCFS:
if self.enable_priority_scheduling: if self.enable_priority_scheduling:
SchedulePolicy._sort_by_priority_and_fcfs( SchedulePolicy._sort_by_priority_and_fcfs(
waiting_queue, self.priority_sign waiting_queue, self.priority_sign
) )
return False return
policy = self._determine_active_policy(waiting_queue) policy = self._determine_active_policy(waiting_queue)
prefix_computed = False
if isinstance(policy, CacheAwarePolicy): if isinstance(policy, CacheAwarePolicy):
prefix_computed = True
temporary_deprioritized = self._compute_prefix_matches( temporary_deprioritized = self._compute_prefix_matches(
waiting_queue, policy waiting_queue, policy
) )
@@ -200,7 +198,6 @@ class SchedulePolicy:
SchedulePolicy._sort_by_routing_key(waiting_queue, running_batch) SchedulePolicy._sort_by_routing_key(waiting_queue, running_batch)
else: else:
raise ValueError(f"Unknown CacheAgnostic Policy: {policy=}") raise ValueError(f"Unknown CacheAgnostic Policy: {policy=}")
return prefix_computed
def _determine_active_policy(self, waiting_queue: List[Req]) -> Policy: def _determine_active_policy(self, waiting_queue: List[Req]) -> Policy:
if self.policy == CacheAwarePolicy.LPM and len(waiting_queue) > 128: if self.policy == CacheAwarePolicy.LPM and len(waiting_queue) > 128: