-
Notifications
You must be signed in to change notification settings - Fork 5.6k
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.
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
Add Bilinear Tensor Product operator. #5014
Changes from 1 commit
611ee68
3ae1424
f5cb52c
4726927
44e1ac3
5cf8204
5f99ae9
ab41648
665eb01
0a6262d
c5d7107
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -43,24 +43,26 @@ class BilinearTensorProductKernel : public framework::OpKernel<T> { | |
|
||
auto batch_size = x->dims()[0]; | ||
auto weight_dims = weight->dims(); | ||
int Out_dim = weight_dims[0]; | ||
int X_dim = weight_dims[1]; | ||
int Y_dim = weight_dims[2]; | ||
auto place = ctx.GetEigenDevice<Place>(); | ||
|
||
// Create the intermediate variable to caculate the result of | ||
// Input(X) multiplied by Input(Weight_i), the formula is: | ||
// left_mul = X Weight_i. | ||
Tensor left_mul; | ||
left_mul.mutable_data<T>(framework::make_ddim({batch_size, weight_dims[2]}), | ||
left_mul.mutable_data<T>(framework::make_ddim({batch_size, Y_dim}), | ||
ctx.GetPlace()); | ||
auto left_mul_mat = EigenMatrix<T>::From(left_mul); | ||
|
||
for (size_t i = 0; i < weight_dims[0]; ++i) { | ||
for (int i = 0; i < Out_dim; ++i) { | ||
auto output_col_vec = output_mat.chip(i, 1); | ||
Tensor weight_mat = weight->Slice(i, i + 1).Resize( | ||
framework::make_ddim({weight_dims[1], weight_dims[2]})); | ||
Tensor weight_mat = | ||
weight->Slice(i, i + 1).Resize(framework::make_ddim({X_dim, Y_dim})); | ||
math::gemm<Place, T>(ctx.device_context(), CblasNoTrans, CblasNoTrans, | ||
batch_size, weight_dims[2], weight_dims[1], 1, | ||
x->data<T>(), weight_mat.data<T>(), 0, | ||
left_mul.data<T>()); | ||
batch_size, Y_dim, X_dim, 1, x->data<T>(), | ||
weight_mat.data<T>(), 0, left_mul.data<T>()); | ||
output_col_vec.device(place) = | ||
(left_mul_mat * y_mat).sum(Eigen::DSizes<int, 1>(1)); | ||
} | ||
|
@@ -87,6 +89,9 @@ class BilinearTensorProductGradKernel : public framework::OpKernel<T> { | |
|
||
auto batch_size = x->dims()[0]; | ||
auto weight_dims = weight->dims(); | ||
int Out_dim = weight_dims[0]; | ||
int X_dim = weight_dims[1]; | ||
int Y_dim = weight_dims[2]; | ||
|
||
auto x_mat = EigenMatrix<T>::From(*x); | ||
auto y_mat = EigenMatrix<T>::From(*y); | ||
|
@@ -95,13 +100,13 @@ class BilinearTensorProductGradKernel : public framework::OpKernel<T> { | |
|
||
// Create the intermediate variable to caculate the Output(Y@Grad). | ||
Tensor x_scale; | ||
x_scale.mutable_data<T>(framework::make_ddim({batch_size, weight_dims[1]}), | ||
x_scale.mutable_data<T>(framework::make_ddim({batch_size, X_dim}), | ||
ctx.GetPlace()); | ||
auto x_scale_mat = EigenMatrix<T>::From(x_scale); | ||
|
||
// Create the intermediate variable to caculate the Output(X@Grad). | ||
Tensor y_scale; | ||
y_scale.mutable_data<T>(framework::make_ddim({batch_size, weight_dims[2]}), | ||
y_scale.mutable_data<T>(framework::make_ddim({batch_size, Y_dim}), | ||
ctx.GetPlace()); | ||
auto y_scale_mat = EigenMatrix<T>::From(y_scale); | ||
|
||
|
@@ -121,51 +126,48 @@ class BilinearTensorProductGradKernel : public framework::OpKernel<T> { | |
|
||
// Caculate the Output(X@Grad) and Output(Y@Grad). | ||
if (d_x || d_y) { | ||
Eigen::DSizes<int, 2> bcast_for_x(1, weight_dims[2]); | ||
Eigen::DSizes<int, 2> bcast_for_y(1, weight_dims[1]); | ||
for (int i = 0; i < weight_dims[0]; ++i) { | ||
Eigen::DSizes<int, 2> bcast_for_x(1, Y_dim); | ||
Eigen::DSizes<int, 2> bcast_for_y(1, X_dim); | ||
for (int i = 0; i < Out_dim; ++i) { | ||
Tensor weight_i = weight->Slice(i, i + 1).Resize( | ||
framework::make_ddim({weight_dims[1], weight_dims[2]})); | ||
framework::make_ddim({X_dim, Y_dim})); | ||
auto output_vec = d_out_mat.chip(i, 1); | ||
if (d_x) { | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. 这里由于broadcast是在batch的方向展开,且TMP = scaled(X) W,scaled(X)中每一行元素所乘的放缩系数不同,所以无法在矩阵乘法之后做scaling计算。即scaled(X) W != scaled(X W). |
||
y_scale_mat.device(place) = | ||
output_vec.reshape(Eigen::DSizes<int, 2>(batch_size, 1)) | ||
.broadcast(bcast_for_x) * | ||
y_mat; | ||
math::gemm<Place, T>(ctx.device_context(), CblasNoTrans, CblasTrans, | ||
batch_size, weight_dims[1], weight_dims[2], 1, | ||
y_scale.data<T>(), weight_i.data<T>(), 1, | ||
d_x->data<T>()); | ||
batch_size, X_dim, Y_dim, 1, y_scale.data<T>(), | ||
weight_i.data<T>(), 1, d_x->data<T>()); | ||
} | ||
if (d_y) { | ||
x_scale_mat.device(place) = | ||
output_vec.reshape(Eigen::DSizes<int, 2>(batch_size, 1)) | ||
.broadcast(bcast_for_y) * | ||
x_mat; | ||
math::gemm<Place, T>(ctx.device_context(), CblasNoTrans, CblasNoTrans, | ||
batch_size, weight_dims[2], weight_dims[1], 1, | ||
x_scale.data<T>(), weight_i.data<T>(), 1, | ||
d_y->data<T>()); | ||
batch_size, Y_dim, X_dim, 1, x_scale.data<T>(), | ||
weight_i.data<T>(), 1, d_y->data<T>()); | ||
} | ||
} | ||
} | ||
|
||
// Caculate the gradient of Input(Weight). | ||
if (d_weight) { | ||
d_weight->mutable_data<T>(ctx.GetPlace()); | ||
Eigen::DSizes<int, 2> bcast_for_weight(1, weight_dims[1]); | ||
for (int i = 0; i < weight_dims[0]; ++i) { | ||
Eigen::DSizes<int, 2> bcast_for_weight(1, X_dim); | ||
for (int i = 0; i < Out_dim; ++i) { | ||
Tensor d_weight_i = d_weight->Slice(i, i + 1).Resize( | ||
framework::make_ddim({weight_dims[1], weight_dims[2]})); | ||
framework::make_ddim({X_dim, Y_dim})); | ||
auto output_vec = d_out_mat.chip(i, 1); | ||
x_scale_mat.device(place) = | ||
output_vec.reshape(Eigen::DSizes<int, 2>(batch_size, 1)) | ||
.broadcast(bcast_for_weight) * | ||
x_mat; | ||
math::gemm<Place, T>(ctx.device_context(), CblasTrans, CblasNoTrans, | ||
weight_dims[1], weight_dims[2], batch_size, 1, | ||
x_scale.data<T>(), y->data<T>(), 0, | ||
d_weight_i.data<T>()); | ||
X_dim, Y_dim, batch_size, 1, x_scale.data<T>(), | ||
y->data<T>(), 0, d_weight_i.data<T>()); | ||
} | ||
} | ||
|
||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Out_dim --> out_dim
X_dim --> x_dim
Y_dim -->y_dim
第一个字母不要大写。
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done