frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core] Refactor: Clean up unused argument in Scheduler._preempt (#9696)

Browse Source 
Signed-off-by: André Jonasson <andre.jonasson@gmail.com>
This commit is contained in:
André Jonasson
2024-11-01 19:41:38 +01:00
committed by GitHub
parent 1dd4cb2935
commit 4581d2cc02
1 changed files with 3 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/core/scheduler.py
View File

@ -828,8 +828,7 @@ class Scheduler:
num_running_seqs)
#Preempt out the victim sequence group
self._preempt(vseq_group, blocks_to_swap_out,
PreemptionMode.RECOMPUTE)
self._preempt(vseq_group, blocks_to_swap_out)
waiting_queue.appendleft(vseq_group)
force_preemption_count += 1
#Put the sequence back into the waiting queue
@ -1451,12 +1450,8 @@ class Scheduler:
if len(cows) > 0:
blocks_to_copy.extend(cows)
def _preempt(
self,
seq_group: SequenceGroup,
blocks_to_swap_out: List[Tuple[int, int]],
preemption_mode: Optional[PreemptionMode] = None,
) -> PreemptionMode:
def _preempt(self, seq_group: SequenceGroup,
blocks_to_swap_out: List[Tuple[int, int]]) -> PreemptionMode:
# If preemption mode is not specified, we determine the mode as follows:
# We use recomputation by default since it incurs lower overhead than
# swapping. However, when the sequence group has multiple sequences