Skip to content

[Layers] fix assert bug when concat gate&up #323

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
abenmao merged 1 commit into from
Apr 18, 2024
Merged

[Layers] fix assert bug when concat gate&up #323

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
3 changes: 3 additions & 0 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -183,6 +183,9 @@ add_definitions(-DAVX512_FP32_WEIGHT_ONLY_NF4=true)
# add_definitions(-DSTEP_BY_STEP_ATTN=true)
add_definitions(-DUSE_SHM=true)
option(XFT_BUILD_TESTS "Build xfastertransformer unit tests" OFF)
if(XFT_BUILD_TESTS)
add_definitions(-DUNDEBUG=true)
Copy link
Contributor

@pujiang2018 pujiang2018 Apr 18, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why do we need UNDEBUG?

Copy link
Contributor Author

@abenmao abenmao Apr 18, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

When we use CMake release, the -DNDEBUG flag is included by default. This option would make all assert statements ineffective as one kind of optimization. I undefine this flag during unit testing (ut) to ensure that assert performs the necessary checks.

Copy link
Contributor Author

@abenmao abenmao Apr 18, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is also why in some environments the assert may fail, while in others it does not. The behavior of assert can vary depending on the compilation settings and the specific conditions.

endif()

# timeline event
option(WITH_TIMELINE "Build with timeline event support" OFF)
Expand Down
10 changes: 6 additions & 4 deletions src/layers/mlp_llama.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,11 @@
// limitations under the License.
// ============================================================================
#pragma once

#ifdef UNDEBUG
#undef NDEBUG
#endif

#include "bert_util.h"
#include "copy_util.h"
#include "debugger.h"
Expand Down Expand Up @@ -248,10 +253,7 @@ class LlamaMLP : public SingletonBase<LlamaMLP<WeiT>> {
TimeLine t("DownProj");

assert(input.Rows() == output.Rows());
if (!enableCATMLP())
assert(input.Cols() == downWeight.Rows());
else
assert(input.Cols() == 2 * downWeight.Rows());
assert(input.Cols() == downWeight.Rows());
assert(downWeight.Cols() == output.Cols());

int M = input.Rows(), N = output.Cols(), K = downWeight.Rows();
Expand Down