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

Minor scheduler fixes (#24359)

Browse Source
This commit is contained in:
Ke Bao
2026-05-05 02:01:23 +08:00
committed by GitHub
parent e6f252e9b8
commit 6dd7aebb36
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/scheduler.py
+2 -3
View File
@@ -2565,8 +2565,7 @@ class Scheduler(
def get_num_allocatable_reqs(self, running_bs): def get_num_allocatable_reqs(self, running_bs):
res = get_global_server_args().pp_max_micro_batch_size - running_bs res = get_global_server_args().pp_max_micro_batch_size - running_bs
if self.pp_size > 1: res = min(res, self.req_to_token_pool.available_size())
res = min(res, self.req_to_token_pool.available_size())
return res return res
def get_new_batch_prefill(self) -> Optional[ScheduleBatch]: def get_new_batch_prefill(self) -> Optional[ScheduleBatch]:
@@ -2599,7 +2598,7 @@ class Scheduler(
if self.enable_hierarchical_cache: if self.enable_hierarchical_cache:
self.tree_cache.check_hicache_events() self.tree_cache.check_hicache_events()
if self.enable_priority_preemption: if self.enable_priority_preemption or self.is_hybrid_swa:
# Reset batch_is_full to try preemption with a prefill adder. # Reset batch_is_full to try preemption with a prefill adder.
self.running_batch.batch_is_full = False self.running_batch.batch_is_full = False