Fix wrong running_bs in priority scheduling (#13142)
Signed-off-by: Tianchen Ding <dtcccc@linux.alibaba.com>
This commit is contained in:
@@ -1806,7 +1806,7 @@ class Scheduler(
|
|||||||
self.running_batch.batch_is_full = True
|
self.running_batch.batch_is_full = True
|
||||||
break
|
break
|
||||||
|
|
||||||
running_bs = len(self.running_batch.reqs) - len(adder.preempt_list)
|
running_bs = len(self.running_batch.reqs)
|
||||||
if len(adder.can_run_list) >= self.get_num_allocatable_reqs(running_bs):
|
if len(adder.can_run_list) >= self.get_num_allocatable_reqs(running_bs):
|
||||||
self.running_batch.batch_is_full = True
|
self.running_batch.batch_is_full = True
|
||||||
if self.disaggregation_mode == DisaggregationMode.PREFILL:
|
if self.disaggregation_mode == DisaggregationMode.PREFILL:
|
||||||
|
|||||||
Reference in New Issue
Block a user