A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[BugFix] fix CUTLASS MLA full cudagraph (#23200)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Lucas Wilkinson committed
Aug 19, 2025 at 18:17 UTC
14e2b0730bf92eb4beaca82e437e3287d6ca82b9
1 file changed
+1
-1
vllm/v1/attention/backends/mla/cutlass_mla.py
+1
-1
@@ -21,7 +21,7 @@ logger = init_logger(__name__)
21
22
class CutlassMLAMetadataBuilder(MLACommonMetadataBuilder[MLACommonMetadata]):
23
# enable full CUDA Graph support for decode-only capture
24
- attn_cudagraph_support: ClassVar[
24
+ cudagraph_support: ClassVar[
25
AttentionCGSupport] = AttentionCGSupport.UNIFORM_SINGLE_TOKEN_DECODE
26
27