Skip to content
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

fix the bug in sigmoid_cross_entropy_with_logits_grad_kernel #64253

Merged
merged 5 commits into from
May 16, 2024
Merged

fix the bug in sigmoid_cross_entropy_with_logits_grad_kernel #64253

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ecd07f4
fix the bug in sigmoid_cross_entropy_with_logits_grad_kernel
zeroRains May 13, 2024
e82f575
add the test for sigmoid_cross_entropy_with_logits_grad
zeroRains May 15, 2024
994fc00
modify the test
zeroRains May 15, 2024
9c55b65
modify the grad kernel
zeroRains May 15, 2024
2b0875d
modify the kernel
zeroRains May 15, 2024
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: 9 additions & 2 deletions paddle/phi/kernels/cpu/sigmoid_cross_entropy_with_logits_grad_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,8 +47,15 @@ void SigmoidCrossEntropyWithLogitsGradKernel(
if (static_cast<int>(label) == ignore_index) {
dx_data[idx] = static_cast<T>(0.);
} else {
T simoid_x = static_cast<T>(1) / (static_cast<T>(1) + std::exp(-x));
T diff = simoid_x * pos_weight_idx - label;
T term1 = (x > 0) ? static_cast<T>(1) : static_cast<T>(0);

T e_x = std::exp(-std::abs(x));
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里为什么要加 std::abs(x) ? e_x 变量不是指的e^(-x) 么?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这是根据前向计算的求导得到的，前向计算是$e^{-|X|}$，其导数是$-e^{-|x|} * where(x>=0,1,0)$

image

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

因为前向中的e^x是e^-abs(x)，所以反向也需要保持e^-abs(x)

T down = 1 + e_x;
T abs_grad = (x >= 0) ? static_cast<T>(1) : static_cast<T>(-1);
T up = -e_x * abs_grad * pos_weight_idx;
T term3 = up / down;

T diff = term1 - label + term3;
dx_data[idx] = dout * diff;
}
}
Expand Down
11 changes: 8 additions & 3 deletions paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,9 +73,14 @@ struct SigmoidBwdPosWeightFunctor {
dx_data = static_cast<T>(0.);
counts = 0;
} else {
T simoid_x =
static_cast<T>(1) / (static_cast<T>(1) + phi::funcs::real_exp(-x));
T diff = simoid_x * pos_weight - label;
T term1 = (x > 0) ? static_cast<T>(1) : static_cast<T>(0);
T e_x = std::exp(-std::abs(x));
T down = 1 + e_x;
T abs_grad = (x >= 0) ? static_cast<T>(1) : static_cast<T>(-1);
T up = -e_x * abs_grad * pos_weight;
T term3 = up / down;

T diff = term1 - label + term3;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

建议调用其他kernel实现

dx_data = dout * diff;
counts = 1;
}
Expand Down