diff --git a/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc b/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc
index 08125ce51b..7b14d496d6 100644
--- a/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc
+++ b/onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc
@@ -72,9 +72,9 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeMatMul);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeLSTM);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, MatMulIntegerToFloat);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearConv);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearConv);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearConv);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, MatMulIntegerToFloat);
-class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t_int8_t, QLinearConv);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, NhwcMaxPool);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, NhwcMaxPool);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, QEmbedLayerNormalization);
@@ -120,7 +120,7 @@ KernelCreateInfo BuildKernelCreateInfo<void>() {
 
 Status RegisterNchwcKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
-      BuildKernelCreateInfo<void>,  //default entry to avoid the list become empty after ops-reducing
+      BuildKernelCreateInfo<void>,  // default entry to avoid the list become empty after ops-reducing
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSNchwcDomain, 1, float, ReorderInput)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSNchwcDomain, 1, float, ReorderOutput)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSNchwcDomain, 1, float, Conv)>,
@@ -143,35 +143,35 @@ Status RegisterNchwcKernels(KernelRegistry& kernel_registry) {
 
 Status RegisterQuantizationKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
-    BuildKernelCreateInfo<void>,  //default entry to avoid the list become empty after ops-reducing
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, MatMulInteger16)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearGlobalAveragePool)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearConcat)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearAveragePool)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, DequantizeLinear)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, DequantizeLinear)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QuantizeLinear)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QuantizeLinear)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearLeakyRelu)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearLeakyRelu)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearSigmoid)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearSigmoid)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearAdd)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearAdd)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearMul)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearMul)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, QAttention)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeMatMul)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeLSTM)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, MatMulIntegerToFloat)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearConv)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, MatMulIntegerToFloat)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t_int8_t, QLinearConv)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, NhwcMaxPool)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, NhwcMaxPool)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, QEmbedLayerNormalization)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QGemm)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QGemm)>,
+      BuildKernelCreateInfo<void>,  // default entry to avoid the list become empty after ops-reducing
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, MatMulInteger16)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearGlobalAveragePool)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearConcat)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, QLinearAveragePool)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, DequantizeLinear)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, DequantizeLinear)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QuantizeLinear)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QuantizeLinear)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearLeakyRelu)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearLeakyRelu)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearSigmoid)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearSigmoid)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearAdd)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearAdd)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearMul)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearMul)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, QAttention)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeMatMul)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, DynamicQuantizeLSTM)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, MatMulIntegerToFloat)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QLinearConv)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QLinearConv)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, MatMulIntegerToFloat)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, NhwcMaxPool)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, NhwcMaxPool)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, QEmbedLayerNormalization)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, int8_t, QGemm)>,
+      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, uint8_t, QGemm)>,
   };
 
   for (auto& function_table_entry : function_table) {
@@ -186,7 +186,7 @@ Status RegisterQuantizationKernels(KernelRegistry& kernel_registry) {
 
 Status RegisterCpuContribKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
-    BuildKernelCreateInfo<void>,  //default entry to avoid the list become empty after ops-reducing
+    BuildKernelCreateInfo<void>,  // default entry to avoid the list become empty after ops-reducing
     BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMSDomain, 1, float, SampleOp)>,
 
     // add more kernels here
diff --git a/onnxruntime/core/providers/cpu/cpu_execution_provider.cc b/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
index ef5e821319..bdf6454285 100644
--- a/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
+++ b/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
@@ -310,13 +310,13 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOn
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, int32_t, DequantizeLinear);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, uint8_t, QuantizeLinear);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, int8_t, QuantizeLinear);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, QLinearMatMul);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t, QLinearMatMul);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t, QLinearMatMul);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t, MatMulInteger);
-class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t_int8_t, QLinearMatMul);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t, MatMulInteger);
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, ConvInteger);
-class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, QLinearConv);
-class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t_int8_t, QLinearConv);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t, QLinearConv);
+class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t, QLinearConv);
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, Slice);
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 11, Dropout);
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10, NonMaxSuppression);
@@ -467,9 +467,9 @@ class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, Ei
 
 // REVIEW(codemzs): ConstEigenVectorArrayMap.cast<MLFLoat16) does not seem to be supported.
 // However these types work on GPU implementation.
-//class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_MLFloat16, Dropout);
-//class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_float, Dropout);
-//class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_double, Dropout);
+// class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_MLFloat16, Dropout);
+// class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_float, Dropout);
+// class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_double, Dropout);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, float_float, Dropout);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, float_double, Dropout);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, double_float, Dropout);
@@ -669,7 +669,7 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, float, Softmax);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, double, Softmax);
 
-//Opset 14
+// Opset 14
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 14, float, CumSum);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 14, double, CumSum);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 14, int32_t, CumSum);
@@ -763,7 +763,7 @@ KernelCreateInfo BuildKernelCreateInfo<void>() {
 
 Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
-    BuildKernelCreateInfo<void>,  //default entry to avoid the list become empty after ops-reducing
+    BuildKernelCreateInfo<void>,  // default entry to avoid the list become empty after ops-reducing
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 6, 10,
                                                                     Clip)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 6, Elu)>,
@@ -1225,15 +1225,15 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
                                                                           QuantizeLinear)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 12, int8_t,
                                                                           QuantizeLinear)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, QLinearMatMul)>,
+    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t, QLinearMatMul)>,
+    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t, QLinearMatMul)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t,
                                                                 MatMulInteger)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t_int8_t, QLinearMatMul)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t,
                                                                 MatMulInteger)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, ConvInteger)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, QLinearConv)>,
-    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t_int8_t, QLinearConv)>,
+    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, uint8_t, QLinearConv)>,
+    BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, int8_t, QLinearConv)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 10,
                                                                     Slice)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, 11,
@@ -1246,7 +1246,7 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
     BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, double,
                                                                 RoiAlign)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 10, ReverseSequence)>,
-    //opset 11
+    // opset 11
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 11,
                                                                     Clip)>,
 
@@ -1515,9 +1515,9 @@ Status RegisterOnnxOperatorKernels(KernelRegistry& kernel_registry) {
     BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, float, Erf)>,
     // REVIEW(codemzs): ConstEigenVectorArrayMap.cast<MLFLoat16) does not seem to be supported.
     // However these types work on GPU implementation.
-    //BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_MLFloat16, Dropout)>,
-    //BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_float, Dropout)>,
-    //BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_double, Dropout)>,
+    // BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_MLFloat16, Dropout)>,
+    // BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_float, Dropout)>,
+    // BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, MLFloat16_double, Dropout)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, float_float, Dropout)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, float_double, Dropout)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 12, 12, double_float, Dropout)>,
@@ -1959,7 +1959,7 @@ KernelCreateInfo BuildKernelCreateInfo<void>() {
 
 Status RegisterOnnxMLOperatorKernels(KernelRegistry& kernel_registry) {
   static const BuildKernelCreateInfoFn function_table[] = {
-      BuildKernelCreateInfo<void>,  //default entry to avoid the list become empty after ops-reducing
+      BuildKernelCreateInfo<void>,  // default entry to avoid the list become empty after ops-reducing
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMLDomain, 1, float,
                                                                   ArrayFeatureExtractor)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kMLDomain, 1, double,
@@ -2073,7 +2073,7 @@ KernelRegistryAndStatus GetCpuKernelRegistry() {
 
 std::shared_ptr<KernelRegistry> CPUExecutionProvider::GetKernelRegistry() const {
   static KernelRegistryAndStatus k = GetCpuKernelRegistry();
-  //throw if the registry failed to initialize
+  // throw if the registry failed to initialize
   ORT_THROW_IF_ERROR(k.st);
   return k.kernel_registry;
 }
diff --git a/onnxruntime/core/providers/cpu/quantization/qlinearconv.cc b/onnxruntime/core/providers/cpu/quantization/qlinearconv.cc
index 68ccb69037..5fa044402d 100644
--- a/onnxruntime/core/providers/cpu/quantization/qlinearconv.cc
+++ b/onnxruntime/core/providers/cpu/quantization/qlinearconv.cc
@@ -15,7 +15,7 @@
 
 //
 // TODO!! Hack! need to move this to MLAS
-// 
+//
 // We use a different job partition with mobile devices
 // where the model tends to be smaller. When the entire
 // weight matrix fits in the cache, we process a thin
@@ -216,7 +216,7 @@ class QLinearConv : public OpKernel {
       packed_W_size_ = MlasSymmQgemmPackBSize(group_output_channels,
                                               kernel_dim,
                                               std::is_same<ActType, int8_t>::value);
-      
+
       if (packed_W_size_ != 0) {
         size_t packed_W_data_size = SafeInt<size_t>(group_count) * packed_W_size_;
         auto* packed_W = static_cast<uint8_t*>(alloc->Alloc(packed_W_data_size));
@@ -285,42 +285,12 @@ class QLinearConv : public OpKernel {
   std::vector<int32_t> column_sums_;
 };
 
-ONNX_CPU_OPERATOR_KERNEL(
+// uint8_t kernel supports weight being either uint8_t or int8_t
+ONNX_OPERATOR_TYPED_KERNEL_EX(
     QLinearConv,
+    kOnnxDomain,
     10,
-    KernelDefBuilder()
-        .TypeConstraint("T1", DataTypeImpl::GetTensorType<uint8_t>())
-        .TypeConstraint("T2", {DataTypeImpl::GetTensorType<uint8_t>(), DataTypeImpl::GetTensorType<int8_t>()})
-        .TypeConstraint("T3", DataTypeImpl::GetTensorType<uint8_t>())
-        .TypeConstraint("T4", DataTypeImpl::GetTensorType<int32_t>()),
-    QLinearConv<uint8_t>);
-
-#define REGISTER_QLINEARCONV_TYPED_KERNEL(domain, version, act_type, weight_type) \
-  ONNX_OPERATOR_TYPED_KERNEL_EX(                                                  \
-      QLinearConv,                                                                \
-      domain,                                                                     \
-      version,                                                                    \
-      act_type##_##weight_type,                                                   \
-      kCpuExecutionProvider,                                                      \
-      KernelDefBuilder()                                                          \
-          .TypeConstraint("T1", DataTypeImpl::GetTensorType<act_type>())          \
-          .TypeConstraint("T2", DataTypeImpl::GetTensorType<weight_type>())       \
-          .TypeConstraint("T3", DataTypeImpl::GetTensorType<act_type>())          \
-          .TypeConstraint("T4", DataTypeImpl::GetTensorType<int32_t>()),          \
-      QLinearConv<act_type>);
-
-REGISTER_QLINEARCONV_TYPED_KERNEL(kOnnxDomain, 10, int8_t, int8_t);
-
-#ifndef DISABLE_CONTRIB_OPS
-
-namespace contrib {
-
-// Register an alternate version of this kernel that supports the channels_last
-// attribute in order to consume and produce NHWC tensors.
-ONNX_OPERATOR_KERNEL_EX(
-    QLinearConv,
-    kMSDomain,
-    1,
+    uint8_t,
     kCpuExecutionProvider,
     KernelDefBuilder()
         .TypeConstraint("T1", DataTypeImpl::GetTensorType<uint8_t>())
@@ -329,7 +299,43 @@ ONNX_OPERATOR_KERNEL_EX(
         .TypeConstraint("T4", DataTypeImpl::GetTensorType<int32_t>()),
     QLinearConv<uint8_t>);
 
-REGISTER_QLINEARCONV_TYPED_KERNEL(kMSDomain, 1, int8_t, int8_t);
+// int8_t kernel only supports weight being int8_t
+#define REGISTER_QLINEARCONV_INT8_KERNEL(domain, version)                \
+  ONNX_OPERATOR_TYPED_KERNEL_EX(                                         \
+      QLinearConv,                                                       \
+      domain,                                                            \
+      version,                                                           \
+      int8_t,                                                            \
+      kCpuExecutionProvider,                                             \
+      KernelDefBuilder()                                                 \
+          .TypeConstraint("T1", DataTypeImpl::GetTensorType<int8_t>())   \
+          .TypeConstraint("T2", DataTypeImpl::GetTensorType<int8_t>())   \
+          .TypeConstraint("T3", DataTypeImpl::GetTensorType<int8_t>())   \
+          .TypeConstraint("T4", DataTypeImpl::GetTensorType<int32_t>()), \
+      QLinearConv<int8_t>);
+
+REGISTER_QLINEARCONV_INT8_KERNEL(kOnnxDomain, 10);
+
+#ifndef DISABLE_CONTRIB_OPS
+
+namespace contrib {
+
+// Register an alternate version of this kernel that supports the channels_last
+// attribute in order to consume and produce NHWC tensors.
+ONNX_OPERATOR_TYPED_KERNEL_EX(
+    QLinearConv,
+    kMSDomain,
+    1,
+    uint8_t,
+    kCpuExecutionProvider,
+    KernelDefBuilder()
+        .TypeConstraint("T1", DataTypeImpl::GetTensorType<uint8_t>())
+        .TypeConstraint("T2", {DataTypeImpl::GetTensorType<uint8_t>(), DataTypeImpl::GetTensorType<int8_t>()})
+        .TypeConstraint("T3", DataTypeImpl::GetTensorType<uint8_t>())
+        .TypeConstraint("T4", DataTypeImpl::GetTensorType<int32_t>()),
+    QLinearConv<uint8_t>);
+
+REGISTER_QLINEARCONV_INT8_KERNEL(kMSDomain, 1);
 
 }  // namespace contrib
 
diff --git a/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc b/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc
index 548a6ac75c..f08bade90a 100644
--- a/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc
+++ b/onnxruntime/core/providers/cpu/quantization/quantize_linear_matmul.cc
@@ -12,10 +12,12 @@
 #include "core/mlas/inc/mlas.h"
 
 namespace onnxruntime {
-ONNX_OPERATOR_KERNEL_EX(
+// uint8_t kernel supports weight being either uint8_t or int8_t
+ONNX_OPERATOR_TYPED_KERNEL_EX(
     QLinearMatMul,
     kOnnxDomain,
     10,
+    uint8_t,
     kCpuExecutionProvider,
     KernelDefBuilder()
         .TypeConstraint("T1", DataTypeImpl::GetTensorType<uint8_t>())
@@ -23,20 +25,21 @@ ONNX_OPERATOR_KERNEL_EX(
         .TypeConstraint("T3", DataTypeImpl::GetTensorType<uint8_t>()),
     QLinearMatMul);
 
-#define REGISTER_QLINEARMATMUL_TYPED_KERNEL(act_type, weight_type)          \
-  ONNX_OPERATOR_TYPED_KERNEL_EX(                                            \
-      QLinearMatMul,                                                        \
-      kOnnxDomain,                                                          \
-      10,                                                                   \
-      act_type##_##weight_type,                                             \
-      kCpuExecutionProvider,                                                \
-      KernelDefBuilder()                                                    \
-          .TypeConstraint("T1", DataTypeImpl::GetTensorType<act_type>())    \
-          .TypeConstraint("T2", DataTypeImpl::GetTensorType<weight_type>()) \
-          .TypeConstraint("T3", DataTypeImpl::GetTensorType<act_type>()),   \
+// int8_t kernel only supports weight being int8_t
+#define REGISTER_QLINEARMATMUL_INT8_KERNEL()                            \
+  ONNX_OPERATOR_TYPED_KERNEL_EX(                                        \
+      QLinearMatMul,                                                    \
+      kOnnxDomain,                                                      \
+      10,                                                               \
+      int8_t,                                                           \
+      kCpuExecutionProvider,                                            \
+      KernelDefBuilder()                                                \
+          .TypeConstraint("T1", DataTypeImpl::GetTensorType<int8_t>())  \
+          .TypeConstraint("T2", DataTypeImpl::GetTensorType<int8_t>())  \
+          .TypeConstraint("T3", DataTypeImpl::GetTensorType<int8_t>()), \
       QLinearMatMul);
 
-REGISTER_QLINEARMATMUL_TYPED_KERNEL(int8_t, int8_t);
+REGISTER_QLINEARMATMUL_INT8_KERNEL();
 
 Status QLinearMatMul::Compute(OpKernelContext* ctx) const {
   const auto* a = ctx->Input<Tensor>(IN_A);
diff --git a/tools/python/util/ort_format_model/operator_type_usage_processors.py b/tools/python/util/ort_format_model/operator_type_usage_processors.py
index 85e5bd1045..92bde6adca 100644
--- a/tools/python/util/ort_format_model/operator_type_usage_processors.py
+++ b/tools/python/util/ort_format_model/operator_type_usage_processors.py
@@ -333,7 +333,7 @@ def _create_operator_type_usage_processors():
     #    ai.onnx: NonMaxSuppression
     #    com.microsoft: FusedConv, FusedGemm, FusedMatMul
     # - Implementation does not have any significant type specific code:
-    #    ai.onnx: Concat, Flatten, Not, QLinearConv, Reshape, Shape, Squeeze, Unsqueeze
+    #    ai.onnx: Concat, Flatten, Not, Reshape, Shape, Squeeze, Unsqueeze
     #
     default_processor_onnx_ops = ['Abs', 'ArgMax', 'ArgMin', 'AveragePool',
                                   'BatchNormalization', 'BitShift',
@@ -346,6 +346,7 @@ def _create_operator_type_usage_processors():
                                   'MatMul', 'Max', 'MaxPool', 'Mean', 'Min',
                                   'NonZero',
                                   'Pad',
+                                  'QLinearConv', 'QLinearMatMul',
                                   'Range', 'Reciprocal', 'ReduceL1', 'ReduceL2', 'ReduceLogSum', 'ReduceLogSumExp',
                                   'ReduceMax', 'ReduceMean', 'ReduceMin', 'ReduceProd', 'ReduceSum', 'ReduceSumSquare',
                                   'Relu', 'Resize', 'ReverseSequence', 'RoiAlign', 'Round',
@@ -365,7 +366,8 @@ def _create_operator_type_usage_processors():
                                                              'Neg',  # used in tflite TransposeConv conversion
                                                              'Sub']
 
-    internal_ops = ['QLinearAdd', 'QLinearMul']
+    # NOTE: QLinearConv has ONNX and internal implementations
+    internal_ops = ['QLinearAdd', 'QLinearMul', 'QLinearConv']
 
     # TODO - review and add ML ops as needed
     # ML Op notes.