xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-06 09:47:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

opt

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2025-08-15 14:24:50 -07:00
parent da03cb8f0b
commit 2ad6985c49
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/v1/attention/backends/flashinfer.py
View File

@ -493,7 +493,8 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
arange = self.block_table_arange[:max_num_blocks].unsqueeze(0)
mask = arange < block_table_bounds_cpu.unsqueeze(1)
if (self.sliding_window is not None and not use_cascade
and num_decodes > 0):
and num_decodes > 0 and
max_num_blocks > self.sliding_window // page_size):
# NOTE(woosuk): Since FlashInfer's decode kernel doesn't skip the kv
# outside the sliding window and only do masking, we manually
# manipulate the seq_lens and block table for skipping.