Skip to content

Gate MoE fusion passes to XeLP+ and align pipeline comments #5

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

Draft
Copilot wants to merge 2 commits into
base: oom/fixing
Choose a base branch
from
Draft

Gate MoE fusion passes to XeLP+ and align pipeline comments #5

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
630c999
Guard MoE fusion passes by XeLP arch
Copilot Jun 2, 2026
9dcc9b0
Clarify comments for XeLP-gated MoE fusion
Copilot Jun 2, 2026
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
5 changes: 3 additions & 2 deletions src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -564,15 +564,16 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {

const bool disable_moe_opt = GPU_DEBUG_VALUE_OR(config.get_disable_moe_opt(), false);

// Run the compressed MoE fusion chain on all devices, including non-systolic iGPU.
// Run compressed MoE conversion passes on all devices, including non-systolic iGPU.
manager.register_pass<ov::pass::ConvertTiledMoeBlockToGatherMatmuls>();

// This pass runs before ConvertPrecision, so f32 activations are still possible here.
// Later MoE fusion passes are applied only on XeLP+.
manager.register_pass<ov::pass::ConvertGatherMatmulToGatherMatmulCompressed>(
std::vector<ov::element::Type>{ov::element::f32, ov::element::f16},
std::vector<ov::element::Type>{ov::element::u4, ov::element::i4,
ov::element::i8, ov::element::u8});
if (!disable_moe_opt) {
if (!disable_moe_opt && device_info.arch >= cldnn::gpu_arch::xe_lp) {
const bool has_batch_dim = !is_pa;
manager.register_pass<ov::pass::MoeOpFusion>(has_batch_dim);
manager.register_pass<ov::intel_gpu::FuseMOESharedExpert>();
Expand Down