xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 20:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core] Early return in SlidingWindowManager.remove_skipped_blocks (#27673)

Browse Source 
Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
This commit is contained in:
Jialin Ouyang 2025-10-28 18:32:33 -07:00 committed by GitHub
parent d3ab240f39
commit 0b51c9bd8b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/core/single_type_kv_cache_manager.py
View File

@ -394,7 +394,13 @@ class SlidingWindowManager(SingleTypeKVCacheManager):
# skipped during the attention computation.
last_useful_token = num_computed_tokens - self.sliding_window + 1
last_useful_block = last_useful_token // self.block_size
if last_useful_block <= 0:
# Early return if tokens are not enough to fill the sliding window
return
blocks = self.req_to_blocks[request_id]
if blocks[last_useful_block - 1] == self._null_block:
# Early return if there are no blocks to remove
return
removed_blocks: list[KVCacheBlock] = []
for i in range(last_useful_block - 1, -1, -1):
if blocks[i] == self._null_block: