squall/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Fix can_swap_in

Browse Source
This commit is contained in:
Woosuk Kwon 2023-02-14 01:57:40 +00:00
parent c574f19506
commit b1644f764f
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
cacheflow/master/block_manager.py
View File

@ -147,12 +147,12 @@ class BlockSpaceManager:
def can_swap_in(self, seq_group: SequenceGroup) -> bool:
blocks = self._get_physical_blocks(seq_group)
num_running_seqs = seq_group.num_seqs(status=SequenceStatus.RUNNING)
num_swapped_seqs = seq_group.num_seqs(status=SequenceStatus.SWAPPED)
num_free_blocks = self.gpu_allocator.get_num_free_blocks()
# NOTE: Conservatively, we assume that every sequence will allocate
# at least one free block right after the swap-in.
# NOTE: This should match the logic in can_append().
return len(blocks) + num_running_seqs <= num_free_blocks
return len(blocks) + num_swapped_seqs <= num_free_blocks
def swap_in(self, seq_group: SequenceGroup) -> Dict[int, int]:
# src_block_number -> dst_block_number