biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

D1.4: Fix regression test for un-normalized O output (D5a)

Browse Source
This commit is contained in:
biondizzle
2026-05-24 15:13:16 +00:00
parent 423d97b094
commit df10378bb5
1 changed files with 23 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
tests/unit/test_d1_regression.py
View File

@@ -1,4 +1,5 @@
"""Quick D1 regression test: HEAD_DIM=64 only, must match Stage C."""
"""Quick D1 regression test: HEAD_DIM=64 only, must match Stage C.
Kernel outputs un-normalized O + LSE (D5a path)."""
import torch, math
import cutlass.cute as cute
import cutlass.torch as ct
@@ -15,31 +16,41 @@ def test():
v_kernel = v.unsqueeze(-1)
c = torch.zeros(m, hd, 1, dtype=torch.bfloat16, device='cuda')
# FP32 reference (un-normalized + normalized)
qf = q[:, :, 0].float()
kf = k[:, :, 0].float()
scale = 1.0 / math.sqrt(hd)
attn = qf @ kf.T * scale
attn = torch.softmax(attn, dim=-1)
ref = attn @ v.float()
attn_max = (qf @ kf.T * scale).max(dim=-1, keepdim=True)[0]
attn_exp = torch.exp(qf @ kf.T * scale - attn_max)
attn_sum = attn_exp.sum(dim=-1, keepdim=True)
ref_unnorm = attn_exp @ v.float()
ref_norm = (attn_exp / attn_sum) @ v.float()
lse_tensor = torch.zeros(m, 1, 1, dtype=torch.float32, device='cuda')
mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q))
mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k))
mV = ct.from_dlpack(v_kernel).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_kernel))
mC = ct.from_dlpack(c).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c))
mLSE = ct.from_dlpack(lse_tensor).mark_layout_dynamic(leading_dim=ct.get_leading_dim(lse_tensor))
stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream)
kernel = FmhaKernel(head_dim=hd, s_k=n)
# normalize=False: kernel outputs un-normalized O + LSE
kernel = FmhaKernel(head_dim=hd, s_k=n, normalize=False)
print(f'hd={hd}, n={n}: Compiling...', flush=True)
compiled = cute.compile(kernel, mQ, mK, mV, mC, stream)
compiled(mQ, mK, mV, mC, stream)
compiled = cute.compile(kernel, mQ, mK, mV, mC, stream, mLSE)
compiled(mQ, mK, mV, mC, stream, mLSE)
torch.cuda.synchronize()
out = c[:, :, 0].float()
cos = torch.nn.functional.cosine_similarity(
out.flatten().unsqueeze(0), ref.flatten().unsqueeze(0)
out_unnorm = c[:, :, 0].float()
out_norm = out_unnorm / attn_sum # external normalization using row_sum
cos_unnorm = torch.nn.functional.cosine_similarity(
out_unnorm.flatten().unsqueeze(0), ref_unnorm.flatten().unsqueeze(0)
).item()
max_abs = (out - ref).abs().max().item()
print(f'hd={hd}, n={n}: cos {cos:.6f} max_abs {max_abs:.4f} {"PASS" if cos >= 0.97 else "FAIL"}')
cos_norm = torch.nn.functional.cosine_similarity(
out_norm.flatten().unsqueeze(0), ref_norm.flatten().unsqueeze(0)
).item()
print(f'hd={hd}, n={n}: cos_unnorm {cos_unnorm:.6f} cos_norm {cos_norm:.6f} {"PASS" if cos_norm >= 0.99 else "FAIL"}')
if __name__ == '__main__':
test()