debug: test 12w identity softmax with n=256 to verify multi-tile pipeline

This commit is contained in:
2026-05-22 10:24:53 +00:00
parent 572656e79b
commit 24a807eae2

View File

@@ -295,7 +295,7 @@ class FmhaV3:
def test():
torch.manual_seed(42)
for n in [128]:
for n in [128, 256]:
m, hd = 128, HEAD_DIM
q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device='cuda')
k = torch.randn(n, hd, 1, dtype=torch.bfloat16, device='cuda')