v4.3 tag release update. (#2789)
This commit is contained in:
@@ -49,9 +49,11 @@ import cutlass.utils.blackwell_helpers as sm100_utils
|
||||
from cutlass.cute.runtime import from_dlpack
|
||||
from cutlass.cute.typing import Int32, Float32, Float8E4M3FN, Float16, BFloat16, Boolean
|
||||
|
||||
current_dir = os.path.dirname(os.path.abspath(__file__))
|
||||
sys.path.insert(0, os.path.join(current_dir, ".."))
|
||||
from utils import fmha_helpers as fmha_utils
|
||||
if __name__ == "__main__":
|
||||
current_dir = os.path.dirname(os.path.abspath(__file__))
|
||||
sys.path.insert(0, os.path.join(current_dir, ".."))
|
||||
|
||||
from helpers import fmha_helpers as fmha_utils
|
||||
|
||||
"""
|
||||
A fused multi-head attention (FMHA) backward pass example for the NVIDIA Blackwell SM100 architecture using CUTE DSL
|
||||
|
||||
Reference in New Issue
Block a user