diff --git a/.vscode/settings.json b/.vscode/settings.json
index 3e2b1f31dd6cf..98d23090fd474 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -21,5 +21,8 @@
     "cpplint.filters": [
         "-build/include_subdir",
         "-runtime/references"
-    ]
+    ],
+    "files.associations": {
+        "span": "cpp"
+    }
 }
diff --git a/onnxruntime/core/optimizer/layer_norm_fusion.cc b/onnxruntime/core/optimizer/layer_norm_fusion.cc
index b6ad4fde6c1f7..ce696154adb6d 100644
--- a/onnxruntime/core/optimizer/layer_norm_fusion.cc
+++ b/onnxruntime/core/optimizer/layer_norm_fusion.cc
@@ -447,6 +447,13 @@ Status LayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int graph_level,
 
     NodeArg* x_input = has_leading_cast ? graph.GetNode(p_reduce_mean_input_node->Index())->MutableInputDefs()[0]
                                         : reduce_mean_node.MutableInputDefs()[0];
+
+    // CPU doesn't support fp16
+    if (reduce_mean_node.GetExecutionProviderType() == kCpuExecutionProvider &&
+        x_input->TypeAsProto()->tensor_type().elem_type() == ONNX_NAMESPACE::TensorProto_DataType_FLOAT16) {
+      continue;
+    }
+
     InlinedVector<NodeArg*> layer_norm_input_defs{x_input, scale, bias};
     Node& layer_norm_node = graph.AddNode(graph.GenerateNodeName("LayerNormalization"),
                                           "LayerNormalization",
@@ -689,6 +696,13 @@ Status SimplifiedLayerNormFusion::ApplyImpl(Graph& graph, bool& modified, int gr
 
     NodeArg* x_input = has_leading_cast ? graph.GetNode(p_pow_input_node->Index())->MutableInputDefs()[0]
                                         : pow_node.MutableInputDefs()[0];
+
+    // CPU doesn't support fp16
+    if (reduce_mean_node.GetExecutionProviderType() == kCpuExecutionProvider &&
+        x_input->TypeAsProto()->tensor_type().elem_type() == ONNX_NAMESPACE::TensorProto_DataType_FLOAT16) {
+      continue;
+    }
+
     InlinedVector<NodeArg*> layer_norm_input_defs{x_input, scale};
     Node& layer_norm_node =
         graph.AddNode(graph.GenerateNodeName("SimplifiedLayerNormalization"), "SimplifiedLayerNormalization",
diff --git a/onnxruntime/core/providers/cpu/cpu_execution_provider.cc b/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
index 37e7e42150413..7e0f919deb0a7 100644
--- a/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
+++ b/onnxruntime/core/providers/cpu/cpu_execution_provider.cc
@@ -143,9 +143,6 @@ class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, Aco
 class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, Atan);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, float, Gemm);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, double, Gemm);
-#ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
-class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8, MLFloat16, Gemm);
-#endif
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 10, Hardmax);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 10, float, LogSoftmax);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 1, 10, double, LogSoftmax);
@@ -335,9 +332,6 @@ class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOn
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 10, Flatten);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 10, float, Gemm);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 10, double, Gemm);
-#ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
-class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 10, MLFloat16, Gemm);
-#endif
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 12, float, MatMul);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 12, double, MatMul);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 12, int32_t, MatMul);
@@ -497,9 +491,6 @@ class ONNX_OPERATOR_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, Sp
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, ScatterND);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, float, Gemm);
 class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, double, Gemm);
-#ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
-class ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, MLFloat16, Gemm);
-#endif
 class ONNX_OPERATOR_VERSIONED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12, GatherElements);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, uint8_t, BitShift);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, uint32_t, BitShift);
@@ -606,9 +597,6 @@ class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain,
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, string, Expand);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, float, Gemm);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, double, Gemm);
-#ifdef MLAS_F16VEC_INTRINSICS_SUPPORTED
-class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, MLFloat16, Gemm);
-#endif
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, float, MatMul);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, double, MatMul);
 class ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, int32_t, MatMul);
@@ -2617,15 +2605,6 @@ Status RegisterFp16Kernels(KernelRegistry& kernel_registry) {
                                                                             MLFloat16, LeakyRelu)>,
       BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 16, MLFloat16,
                                                                   LeakyRelu)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 7, 8,
-                                                                            MLFloat16, Gemm)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 9, 10,
-                                                                            MLFloat16, Gemm)>,
-      BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 11, 12,
-                                                                            MLFloat16, Gemm)>,
-
-      BuildKernelCreateInfo<ONNX_OPERATOR_TYPED_KERNEL_CLASS_NAME(kCpuExecutionProvider, kOnnxDomain, 13, MLFloat16,
-                                                                  Gemm)>,
   };
 
   for (auto& function_table_entry : function_table) {
diff --git a/onnxruntime/test/providers/cpu/math/gemm_test.cc b/onnxruntime/test/providers/cpu/math/gemm_test.cc
index 428925e154497..1a542fb67418e 100644
--- a/onnxruntime/test/providers/cpu/math/gemm_test.cc
+++ b/onnxruntime/test/providers/cpu/math/gemm_test.cc
@@ -277,7 +277,7 @@ class GemmOpTypedTests : public ::testing::Test {
 // On CPUs without fp16 instructions the tests will output a warning:
 // "registered execution providers CPUExecutionProvider were unable to run the model"
 // , then they will still pass.
-using GemmOpTypedTestsTypes = ::testing::Types<float, double, MLFloat16>;
+using GemmOpTypedTestsTypes = ::testing::Types<float, double>;
 TYPED_TEST_SUITE(GemmOpTypedTests, GemmOpTypedTestsTypes);
 
 TYPED_TEST(GemmOpTypedTests, TestGemmScalarBroadcast) {