Flashinfer cuDNN backend for Qwen3 VL ViT attention (#34580)

Signed-off-by: Max Hu <maxhu@nvidia.com>
Signed-off-by: Max Hu <hyoung2991@gmail.com>
Co-authored-by: Max Hu <maxhu@nvidia.com>
Co-authored-by: Shang Wang <shangw@nvidia.com>
This commit is contained in:
Max Hu
2026-02-27 20:20:23 +08:00
committed by GitHub
co-authored by Max Hu Shang Wang
parent b66a74649e
commit 9c3fe9936b
6 changed files with 405 additions and 21 deletions
+1
View File
@@ -414,6 +414,7 @@ class CudaPlatformBase(Platform):
AttentionBackendEnum.FLASH_ATTN,
AttentionBackendEnum.TRITON_ATTN,
AttentionBackendEnum.TORCH_SDPA,
AttentionBackendEnum.FLASHINFER,
]
@classmethod