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

[CodeStyle][Typos][N-11] Fix typos (Normlized, normlized) #70263

Merged
merged 1 commit into from
Dec 17, 2024
Merged

[CodeStyle][Typos][N-11] Fix typos (Normlized, normlized) #70263

Show file tree
Hide file tree
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
2 changes: 0 additions & 2 deletions _typos.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,8 +346,6 @@ neigbhors = 'neigbhors'
Neigbors = 'Neigbors'
neighor = 'neighor'
netwrok = 'netwrok'
normlized = 'normlized'
Normlized = 'Normlized'
normlize = 'normlize'
noraml = 'noraml'
numer = 'numer'
Expand Down
18 changes: 9 additions & 9 deletions paddle/fluid/primitive/decomp_rule/decomp_vjp/details.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -834,16 +834,16 @@ void layer_norm_grad(const Tensor& x,
auto bias_ptr = bias.get_ptr();
LayerNormDecompHelper decomp_help(x, scale, bias, begin_norm_axis);

std::vector<int64_t> normlized_axis;
std::vector<int64_t> normalized_axis;
std::vector<int64_t> mean_var_new_shape(mean.dims().size(), 0);
for (int i = begin_norm_axis; i < x_dims.size(); ++i) {
mean_var_new_shape.push_back(1);
normlized_axis.push_back(i);
normalized_axis.push_back(i);
}

std::vector<int64_t> un_normlized_axis;
std::vector<int64_t> un_normalized_axis;
for (int i = 0; i < begin_norm_axis; ++i) {
un_normlized_axis.push_back(i);
un_normalized_axis.push_back(i);
}

auto mean_ = reshape<T>(mean, mean_var_new_shape);
Expand Down Expand Up @@ -875,14 +875,14 @@ void layer_norm_grad(const Tensor& x,
}

auto dx_end = sqrt_var_1 * out_grad_scale;
auto d_mean = dx_end.sum(normlized_axis, x_cast.dtype(), true); // M,1
auto d_mean = dx_end.sum(normalized_axis, x_cast.dtype(), true); // M,1

auto d_std_1 = (tmp * x_sub_mean * out_grad_scale)
.sum(normlized_axis, x_cast.dtype(), true); // M,1
.sum(normalized_axis, x_cast.dtype(), true); // M,1
auto d_std = d_std_1 * x_sub_mean_mul_sqrt_var_1; // M,1 * M,N = M,N

auto d_mean_d_std =
(d_mean + d_std) / decomp_help.GetNormlizedNumel<T>(d_std);
(d_mean + d_std) / decomp_help.GetNormalizedNumel<T>(d_std);

auto x_grad_tmp = dx_end - d_mean_d_std;
x_grad_tmp = ConverToOrig<T>(x_grad_tmp, x.dtype());
Expand All @@ -893,7 +893,7 @@ void layer_norm_grad(const Tensor& x,
if (scale_grad) {
if (scale_ptr) {
auto scale_grad_tmp = (x_sub_mean_mul_sqrt_var_1 * out_grad_cast)
.sum(un_normlized_axis, x_cast.dtype(), true);
.sum(un_normalized_axis, x_cast.dtype(), true);
scale_grad_tmp = reshape<T>(scale_grad_tmp, {-1});
scale_grad_tmp = ConverToOrig<T>(scale_grad_tmp, scale_ptr->dtype());

Expand All @@ -906,7 +906,7 @@ void layer_norm_grad(const Tensor& x,
if (bias_grad) {
if (bias_ptr) {
auto bias_grad_tmp =
out_grad_cast.sum(un_normlized_axis, x_cast.dtype(), true);
out_grad_cast.sum(un_normalized_axis, x_cast.dtype(), true);
bias_grad_tmp = reshape<T>(bias_grad_tmp, {-1});
bias_grad_tmp = ConverToOrig<T>(bias_grad_tmp, bias_ptr->dtype());

Expand Down
24 changes: 12 additions & 12 deletions paddle/fluid/primitive/decomp_utils/decomp_utils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,22 +322,22 @@ class LayerNormDecompHelper {
for (int i = begin_norm_axis; i < x_rank_; ++i) {
if (x_dims[i] < 0) {
static_norm_shape_ = false;
normlized_numel_ = -1;
normalized_numel_ = -1;
break;
}

normlized_shape_.push_back(x_dims[i]);
normalized_shape_.push_back(x_dims[i]);

normlized_numel_ *= x_dims[i];
normalized_numel_ *= x_dims[i];
}

if (!static_norm_shape_) {
// try get static norm numel from sacle for bias
normlized_numel_ = -1;
normalized_numel_ = -1;
if (scale.get_ptr()) {
normlized_numel_ = scale->dims()[0];
normalized_numel_ = scale->dims()[0];
} else if (bias.get_ptr()) {
normlized_numel_ = bias->dims()[0];
normalized_numel_ = bias->dims()[0];
}
}
}
Expand All @@ -349,17 +349,17 @@ class LayerNormDecompHelper {
}

if (static_norm_shape_) {
return reshape<T>(s, normlized_shape_);
return reshape<T>(s, normalized_shape_);
} else {
return backend::reshape<T>(
s, get_slice_vec<T>(shape64<T>(x), begin_norm_axis_, x_rank_));
}
}

template <typename T>
Tensor GetNormlizedNumel(const Tensor& x) {
if (normlized_numel_ != -1) {
return full_scalar<T>(normlized_numel_, x.dtype());
Tensor GetNormalizedNumel(const Tensor& x) {
if (normalized_numel_ != -1) {
return full_scalar<T>(normalized_numel_, x.dtype());
} else {
auto x_shape = shape64<T>(x);
auto numel = get_slice<T>(x_shape, begin_norm_axis_);
Expand All @@ -372,11 +372,11 @@ class LayerNormDecompHelper {
}

private:
std::vector<int64_t> normlized_shape_;
std::vector<int64_t> normalized_shape_;
bool scale_need_reshape_;
bool static_norm_shape_;
int64_t x_rank_;
int64_t normlized_numel_{1};
int64_t normalized_numel_{1};
int begin_norm_axis_;
};

Expand Down