Skip to content

Update QMoE kernel with optimizations #26091

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

Open
apsonawane wants to merge 4 commits into
base: main
Choose a base branch
from
+141 −337
Open

Update QMoE kernel with optimizations #26091

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
83ffc5d
Fix merge conflicts
apsonawane Sep 25, 2025
d399c66
Re-enable quantized Mlas
apsonawane Sep 25, 2025
8289fcb
Add overflow safety changes
apsonawane Sep 25, 2025
d57a7c3
Disable quantized Mlas, still not giving good tps
apsonawane Sep 26, 2025
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
14 changes: 14 additions & 0 deletions onnxruntime/contrib_ops/cpu/moe/moe_helper.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,8 +33,22 @@ struct MoEParameters {
MoEParallelType parallel_type{MoEParallelType::None};
int64_t tensor_shards{1};
};

namespace moe_helper {

// Validate block-wise quantization requirements according to op spec
inline Status ValidateBlockwiseQuantization(int64_t block_size, int64_t hidden_size, int64_t inter_size) {
if (block_size > 0) {
ORT_ENFORCE(hidden_size % block_size == 0,
"For block-wise quantization, hidden_size (", hidden_size,
") must be divisible by block_size (", block_size, ").");
ORT_ENFORCE(inter_size % block_size == 0,
"For block-wise quantization, inter_size (", inter_size,
") must be divisible by block_size (", block_size, ").");
}
return Status::OK();
}

template <typename Tensor>
Status CheckInputs(MoEParameters& parameters,
const Tensor* input, // required
Expand Down
Loading
Loading