diff --git a/python/sglang/srt/managers/scheduler.py b/python/sglang/srt/managers/scheduler.py index 1d4b9fd14..de2226f66 100644 --- a/python/sglang/srt/managers/scheduler.py +++ b/python/sglang/srt/managers/scheduler.py @@ -2565,8 +2565,7 @@ class Scheduler( def get_num_allocatable_reqs(self, 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 def get_new_batch_prefill(self) -> Optional[ScheduleBatch]: @@ -2599,7 +2598,7 @@ class Scheduler( if self.enable_hierarchical_cache: 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. self.running_batch.batch_is_full = False