Skip to content

Fix bias shape in weightOnlyGroupwiseQuantMatmulPlugin for TRT workflow #4348

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
May 16, 2025
Merged

Fix bias shape in weightOnlyGroupwiseQuantMatmulPlugin for TRT workflow #4348

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
8 changes: 4 additions & 4 deletions ...llm/plugins/weightOnlyGroupwiseQuantMatmulPlugin/weightOnlyGroupwiseQuantMatmulPlugin.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ void WeightOnlyGroupwiseQuantGemmPluginProfiler::runTactic(int m, int n, int k,
nextWorkspacePtr(reinterpret_cast<int8_t*>(inputScalesPtr), k * originalN * sizeof(half) / mGroupSize));
half* biasesPtr = reinterpret_cast<half*>(
nextWorkspacePtr(reinterpret_cast<int8_t*>(zerosPtr), k * originalN * sizeof(half) / mGroupSize));
half* outputPtr = reinterpret_cast<half*>(nextWorkspacePtr(reinterpret_cast<int8_t*>(biasesPtr), m * sizeof(half)));
half* outputPtr = reinterpret_cast<half*>(nextWorkspacePtr(reinterpret_cast<int8_t*>(biasesPtr), n * sizeof(half)));
char* workspacePtr
= reinterpret_cast<char*>(nextWorkspacePtr(reinterpret_cast<int8_t*>(outputPtr), m * originalN * sizeof(half)));
if ((mQuantAlgo & GroupwiseQuantAlgo::ZERO) == 0)
Expand Down Expand Up @@ -95,7 +95,7 @@ void WeightOnlyGroupwiseQuantGemmPluginProfiler::computeTmpSize(size_t maxM, siz
k * n * sizeof(float), // B
k * originalN * sizeof(half) / mGroupSize, // scales
k * originalN * sizeof(half) / mGroupSize, // zeros
maxM * sizeof(half), // biases
originalN * sizeof(half), // biases
maxM * originalN * sizeof(half), // C
mRunner->getWorkspaceSize(maxM, originalN, k) // workspace
};
Expand Down Expand Up @@ -309,7 +309,7 @@ nvinfer1::DimsExprs WeightOnlyGroupwiseQuantMatmulPlugin::getOutputDimensions(
// 2 weights [K, N/2]
// 3 scales [K // group_size, N]
// 4 zeros [K // group_size, N] (optional)
// 5 biases [M] (optional)
// 5 biases [N] (optional)

try
{
Expand Down Expand Up @@ -414,7 +414,7 @@ int WeightOnlyGroupwiseQuantMatmulPlugin::enqueue(nvinfer1::PluginTensorDesc con
// 2 weights [K, N/2]
// 3 scales [K // group_size, N]
// 4 zeros [K // group_size, N]
// 5 biases [M]
// 5 biases [N]
// outputs
// mat [M, N]

Expand Down