xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 06:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Attention] Remove max cudagraph size limit of 992 (#27840)

Browse Source 
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
22quinn 2025-11-07 22:33:27 -08:00 committed by GitHub
parent 4a36681f85
commit 608bb14462
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 0 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/v1/attention/backends/flash_attn.py
View File

@ -244,13 +244,6 @@ class FlashAttentionMetadataBuilder(AttentionMetadataBuilder[FlashAttentionMetad
self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size
if self.use_full_cuda_graph and self.aot_schedule: if self.use_full_cuda_graph and self.aot_schedule:
if self.max_cudagraph_size > 992:
# This condition derives from FA3's internal heuristic.
# TODO(woosuk): Support larger cudagraph sizes.
raise ValueError(
"Capture size larger than 992 is not supported for full cuda graph."
)
self.scheduler_metadata = torch.zeros( self.scheduler_metadata = torch.zeros(
vllm_config.scheduler_config.max_num_seqs + 1, vllm_config.scheduler_config.max_num_seqs + 1,
dtype=torch.int32, dtype=torch.int32,

7
vllm/v1/attention/backends/mla/flashattn_mla.py
View File

@ -97,13 +97,6 @@ class FlashAttnMLAMetadataBuilder(MLACommonMetadataBuilder[FlashAttnMLAMetadata]
self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size
if self.use_full_cuda_graph and self.fa_aot_schedule: if self.use_full_cuda_graph and self.fa_aot_schedule:
if self.max_cudagraph_size > 992:
# This condition derives from FA3's internal heuristic.
# TODO(woosuk): Support larger cudagraph sizes.
raise ValueError(
"Capture size larger than 992 is not supported for full cuda graph."
)
self.scheduler_metadata = torch.zeros( self.scheduler_metadata = torch.zeros(
vllm_config.scheduler_config.max_num_seqs + 1, vllm_config.scheduler_config.max_num_seqs + 1,
dtype=torch.int32, dtype=torch.int32,