Skip to content

Detect Hopper for FlashAttention 3 by compute capability, not GPU name (H200/H800/H20 silently fell back to FA2) #13

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
frankjiang merged 1 commit into from
Sep 14, 2026
Merged

Detect Hopper for FlashAttention 3 by compute capability, not GPU name (H200/H800/H20 silently fell back to FA2) #13

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions wan/modules/attention.py
100755 → 100644
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,11 +5,16 @@
import flash_attn_interface

def is_hopper_gpu():
"""flashattn-hopper 仅支持 Hopper (H100)，不支持 Blackwell，故只检测 Hopper。"""
"""flashattn-hopper targets sm_90 only (Hopper, not Blackwell).

Detect by compute capability rather than by product name: H200, H800
and H20 are sm_90 too, but their names contain neither "h100" nor
"hopper", so a name check silently routed them to FlashAttention 2.
"""
if not torch.cuda.is_available():
return False
device_name = torch.cuda.get_device_name(0).lower()
return "h100" in device_name or "hopper" in device_name
major, _ = torch.cuda.get_device_capability(0)
return major == 9
FLASH_ATTN_3_AVAILABLE = is_hopper_gpu()
except ModuleNotFoundError:
FLASH_ATTN_3_AVAILABLE = False
Expand Down