From 4b028e619990c232cf6730bb47a8de4e967f507b Mon Sep 17 00:00:00 2001
From: James Bartlett <james@gimletlabs.ai>
Date: Wed, 4 Dec 2024 14:45:27 -0800
Subject: [PATCH] [GML] Fix quantization pipelines

Signed-off-by: James Bartlett <james@gimletlabs.ai>
---
 .../src/transformations/transformation_pipeline.cpp         | 2 +-
 .../intel_gpu/src/plugin/transformations_pipeline.cpp       | 6 +++---
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
index 9dd1da2d471e5a..f3acc8bf0f671e 100644
--- a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
+++ b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp
@@ -683,7 +683,7 @@ void Transformations::PreLpt(const std::vector<ov::element::Type>& defaultPrecis
 
         CPU_SET_CALLBACK_COMMON(manager,
             [&defaultPrecisions](const_node_ptr &node) -> bool {
-                return ov::pass::low_precision::NetworkHelper::areQuantizeAndDequantizeSupportedForMultiply(node, defaultPrecisions);
+                return !ov::pass::low_precision::NetworkHelper::areQuantizeAndDequantizeSupportedForMultiply(node, defaultPrecisions);
             },
             ov::pass::ConvertQuantizeDequantize);
     }
diff --git a/src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp b/src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
index 158dee2ee7ac05..63ceebfd993621 100644
--- a/src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
+++ b/src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
@@ -338,6 +338,8 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
             });
         }
 
+        manager.register_pass<ov::pass::CommonOptimizations>();
+
         const bool keep_precision_sensitive_in_fp32_1 = true;
         const bool convert_input_output_precision = false;
         const bool store_original_precision_as_rt_attribute = true;
@@ -347,8 +349,6 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
                                                           convert_input_output_precision,
                                                           store_original_precision_as_rt_attribute);
 
-        manager.register_pass<ov::pass::CommonOptimizations>();
-
         pass_config->set_callback<ov::pass::ScaledDotProductAttentionDecomposition>([&](const std::shared_ptr<const ov::Node> node){
             GPU_DEBUG_IF(cldnn::debug_configuration::get_instance()->enable_sdpa != -1) {
                 GPU_DEBUG_CODE(return cldnn::debug_configuration::get_instance()->enable_sdpa == 1);
@@ -667,7 +667,7 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
 
         if (enableInt8) {
             pass_config->set_callback<ov::pass::ConvertQuantizeDequantize>([&](const_node_ptr &node) -> bool {
-                return ov::pass::low_precision::NetworkHelper::areQuantizeAndDequantizeSupportedForMultiply(node, defaultPrecisions);
+                return !ov::pass::low_precision::NetworkHelper::areQuantizeAndDequantizeSupportedForMultiply(node, defaultPrecisions);
             });
         }