xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-08 21:15:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix a ref count bug in swapping

Browse Source
This commit is contained in:
Woosuk Kwon 2023-02-14 09:20:12 +00:00
parent c128c2ed03
commit 2729087efe
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cacheflow/master/block_manager.py
View File

@ -166,6 +166,7 @@ class BlockSpaceManager:
for cpu_block in block_table: for cpu_block in block_table:
if cpu_block in mapping: if cpu_block in mapping:
gpu_block = mapping[cpu_block] gpu_block = mapping[cpu_block]
gpu_block.ref_count += 1
else: else:
gpu_block = self.gpu_allocator.allocate() gpu_block = self.gpu_allocator.allocate()
mapping[cpu_block] = gpu_block mapping[cpu_block] = gpu_block
@ -196,6 +197,7 @@ class BlockSpaceManager:
for gpu_block in block_table: for gpu_block in block_table:
if gpu_block in mapping: if gpu_block in mapping:
cpu_block = mapping[gpu_block] cpu_block = mapping[gpu_block]
cpu_block.ref_count += 1
else: else:
cpu_block = self.cpu_allocator.allocate() cpu_block = self.cpu_allocator.allocate()
mapping[gpu_block] = cpu_block mapping[gpu_block] = cpu_block