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

[TEST][GPU][QWen3] WA FC fusion + RMS issue #29759

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
+2 −2
Open

[TEST][GPU][QWen3] WA FC fusion + RMS issue #29759

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
2 changes: 1 addition & 1 deletion src/plugins/intel_gpu/include/intel_gpu/runtime/options.inl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,7 @@ OV_CONFIG_RELEASE_OPTION(ov::hint, activations_scale_factor, -1.0f, "Scalar floa
OV_CONFIG_RELEASE_OPTION(ov::internal, enable_lp_transformations, false, "Enable/Disable Low precision transformations set")
OV_CONFIG_RELEASE_OPTION(ov::intel_gpu, config_file, "", "Path to custom layers config file")
OV_CONFIG_RELEASE_OPTION(ov::hint, model, nullptr, "Shared pointer to the ov::Model")
OV_CONFIG_RELEASE_OPTION(ov::intel_gpu, disable_horizontal_fc_fusion, true, "Disable pass which merges QKV projections into single MatMul")

OV_CONFIG_RELEASE_INTERNAL_OPTION(ov::intel_gpu, shape_predictor_settings, {10, 16 * 1024, 2, 1.1f}, "Preallocation settings")
OV_CONFIG_RELEASE_INTERNAL_OPTION(ov::intel_gpu, queue_type, QueueTypes::out_of_order, "Type of the queue that must be used for model execution. May be in-order or out-of-order")
Expand Down Expand Up @@ -73,7 +74,6 @@ OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, host_time_profiling, 0, "Measre and print
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_async_compilation, false, "Disable feature that allows to asyncrhonously prepare static-shaped implementations for the primitives with shape-agnostic kernels selected during compilation")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_runtime_buffer_fusing, false, "Disable runtime inplace optimizations for operations like concat and crop")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_post_ops_fusions, false, "Disable fusions of operations as post-ops/fused-ops")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_horizontal_fc_fusion, false, "Disable pass which merges QKV projections into single MatMul")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_fc_swiglu_fusion, false, "Disable pass which merges FC and SwiGLU ops")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_fake_alignment, false, "Disable fake alignment feature which tries to keep gpu friendly memory alignment for arbitrary tensor shapes")
OV_CONFIG_DEBUG_OPTION(ov::intel_gpu, disable_memory_reuse, false, "Disable memory reuse for activation tensors")
Expand Down
2 changes: 1 addition & 1 deletion 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 @@ -1106,7 +1106,7 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
manager.register_pass<ov::intel_gpu::MoveFCReshapeToWeights>();
manager.register_pass<ov::intel_gpu::ConvertFullyConnectedToFullyConnectedCompressed>();

bool disable_horizontal_fc_fusion = GPU_DEBUG_VALUE_OR(config.get_disable_horizontal_fc_fusion(), false);
bool disable_horizontal_fc_fusion = GPU_DEBUG_VALUE_OR(config.get_disable_horizontal_fc_fusion(), true);
bool disable_fc_swiglu_fusion = GPU_DEBUG_VALUE_OR(config.get_disable_fc_swiglu_fusion(), false);

// mlp fusion is only supported for cldnn on high performant GPUis
Expand Down
Loading