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 GroupNorm backward grad #10045

Merged
merged 8 commits into from
Mar 29, 2023
Merged

fix GroupNorm backward grad #10045

Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
b8995bb
execute group_norm_param_grad only when gamma and beta require grad
fpzh2011 Mar 27, 2023
971f60c
refine group_norm_param grad
fpzh2011 Mar 27, 2023
d9f15ea
Merge branch 'master' into fix_group_norm_grad
fpzh2011 Mar 28, 2023
e132889
auto format by CI
oneflow-ci-bot Mar 28, 2023
5a67951
Merge branch 'master' into fix_group_norm_grad
mergify[bot] Mar 28, 2023
4bc271f
Merge branch 'master' into fix_group_norm_grad
fpzh2011 Mar 29, 2023
6544c10
Merge branch 'fix_group_norm_grad' of github.com:Oneflow-Inc/oneflow …
fpzh2011 Mar 29, 2023
38b0c5b
Merge branch 'master' into fix_group_norm_grad
fpzh2011 Mar 29, 2023
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
10 changes: 7 additions & 3 deletions oneflow/core/autograd/gradient_funcs/group_norm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,8 @@ namespace one {
struct GroupNormCaptureState : public AutoGradCaptureState {
double epsilon = 1e-5;
bool x_requires_grad = true;
bool gamma_requires_grad = true;
bool beta_requires_grad = true;
bool affine = true;
int32_t num_groups = 1;
size_t x_index = 0;
Expand Down Expand Up @@ -66,6 +68,8 @@ Maybe<void> GroupNorm::Capture(GroupNormCaptureState* ctx, const TensorTuple& in
ctx->activation = JUST(composed_attrs.GetAttr<std::string>("activation"));
if (ctx->affine) {
CHECK_EQ_OR_RETURN(inputs.size(), 3); // NOLINT(maybe-need-error-msg)
ctx->gamma_requires_grad = inputs.at(1)->requires_grad();
ctx->beta_requires_grad = inputs.at(2)->requires_grad();
} else {
CHECK_EQ_OR_RETURN(inputs.size(), 1); // NOLINT(maybe-need-error-msg)
}
Expand Down Expand Up @@ -98,10 +102,10 @@ Maybe<void> GroupNorm::Apply(const GroupNormCaptureState* ctx, const TensorTuple
const auto& mean = saved_tensors.at(ctx->mean_index);
const auto& inv_variance = saved_tensors.at(ctx->inv_variance_index);

if (ctx->affine) {
if (ctx->affine && (ctx->gamma_requires_grad || ctx->beta_requires_grad)) {
const auto& results = JUST(functional::GroupNormParamGrad(dy, x, mean, inv_variance));
in_grads->at(1) = results->at(0); // For gamma.
in_grads->at(2) = results->at(1); // For beta.
if (ctx->gamma_requires_grad) { in_grads->at(1) = results->at(0); } // For gamma.
if (ctx->beta_requires_grad) { in_grads->at(2) = results->at(1); } // For beta.
}
if (ctx->x_requires_grad) {
if (ctx->affine) {
Expand Down