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

Jump to bottom

Add xentropy bf16 support #1790

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Add xentropy bf16 support #1790

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
4 changes: 2 additions & 2 deletions apex/contrib/csrc/xentropy/xentropy_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -574,7 +574,7 @@ std::vector<Tensor> host_softmax_xentropy(
const Tensor & labels_,
const float smoothing,
const bool half_to_float){
if (half_to_float) TORCH_CHECK(input_.scalar_type() == ScalarType::Half,"conversion is supported for Half type only");
if (half_to_float) TORCH_CHECK(input_.scalar_type() == ScalarType::Half || input_.scalar_type() == ScalarType::BFloat16,"conversion is supported for Half type only");
TORCH_CHECK(labels_.scalar_type() == ScalarType::Long,"Label type should be CUDA Long");

auto input = input_.contiguous();
Expand Down Expand Up @@ -712,7 +712,7 @@ at::Tensor softmax_xentropy_backward_cuda(
const float smoothing) {
bool half_to_float = grad_loss.scalar_type() != logits.scalar_type();
if (half_to_float) {
TORCH_CHECK((grad_loss.scalar_type() == ScalarType::Float && logits.scalar_type() == ScalarType::Half), "expected input and grad types to match, or input to be at::Half and grad to be at::Float");
TORCH_CHECK((grad_loss.scalar_type() == ScalarType::Float && (logits.scalar_type() == ScalarType::Half || logits.scalar_type() == ScalarType::BFloat16)), "expected input and grad types to match, or input to be at::Half and grad to be at::Float");
}
return host_softmax_xentropy_backward<LogSoftMaxBackwardEpilogue>(grad_loss, logits, max_log_sum_exp, labels, smoothing, half_to_float);
}