Skip to content

fix ignore_index label check in softmax_with_cross_entropy #31201

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

Merged
merged 1 commit into from
Feb 25, 2021
Merged

fix ignore_index label check in softmax_with_cross_entropy #31201

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
15 changes: 9 additions & 6 deletions paddle/fluid/operators/softmax_with_cross_entropy_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -253,12 +253,13 @@ struct HardLabelSoftmaxWithCrossEntropyFunctor {
public:
HardLabelSoftmaxWithCrossEntropyFunctor(const int64_t* labels, T* loss,
T* log_softmax, int64_t d,
int axis_dim)
int axis_dim, int ignore_idx)
: labels_(labels),
loss_(loss),
log_softmax_(log_softmax),
d_(d),
axis_dim_(axis_dim) {}
axis_dim_(axis_dim),
ignore_idx_(ignore_idx) {}

__device__ void operator()(int64_t idx) const {
// logits view as [n, axis_dim, remain], where d = axis_dim * remain
Expand All @@ -268,10 +269,11 @@ struct HardLabelSoftmaxWithCrossEntropyFunctor {
int64_t idx_remain = idx % remain;
// labels, loss view as [n, remain]
int64_t idx_lbl = idx_n * remain + idx_remain;
PADDLE_ENFORCE(labels_[idx_lbl] >= 0 && labels_[idx_lbl] < d_,
"The value of label[%ld] expected >= 0 and < %ld,"
PADDLE_ENFORCE(labels_[idx_lbl] >= 0 && labels_[idx_lbl] < d_ ||
labels_[idx_lbl] == ignore_idx_,
"The value of label[%ld] expected >= 0 and < %ld, or == %d,"
"but got %ld. Please check input value.",
idx_lbl, d_, labels_[idx_lbl]);
idx_lbl, d_, ignore_idx_, labels_[idx_lbl]);
// It also would ignore labels not in range(class_num).
if (idx_axis != labels_[idx_lbl]) {
log_softmax_[idx] = exp_on_device(log_softmax_[idx]);
Expand All @@ -288,6 +290,7 @@ struct HardLabelSoftmaxWithCrossEntropyFunctor {
T* log_softmax_;
int64_t d_;
int axis_dim_;
int ignore_idx_;
};

template <typename T>
Expand Down Expand Up @@ -355,7 +358,7 @@ static void HardLabelSoftmaxWithCrossEntropy(
labels_data, loss_data, softmax_data, d, axis_dim, ignore_idx)); \
} else { \
for_range(HardLabelSoftmaxWithCrossEntropyFunctor<T>( \
labels_data, loss_data, softmax_data, d, axis_dim)); \
labels_data, loss_data, softmax_data, d, axis_dim, ignore_idx)); \
} \
} break

Expand Down