diff --git a/backends/cadence/fusion_g3/operators/op_dequantize.cpp b/backends/cadence/fusion_g3/operators/op_dequantize.cpp
index 53478ba4626..0e92f2158fd 100644
--- a/backends/cadence/fusion_g3/operators/op_dequantize.cpp
+++ b/backends/cadence/fusion_g3/operators/op_dequantize.cpp
@@ -11,7 +11,6 @@
 #include <algorithm>
 #include <cinttypes>
 #include <cmath>
-#include <optional>
 
 #include <xa_nnlib_kernels_api.h>
 
@@ -23,6 +22,8 @@ using ::executorch::aten::ScalarType;
 using ::executorch::aten::Tensor;
 using ::executorch::runtime::KernelRuntimeContext;
 
+template <typename T>
+using optional = std::optional<T>;
 /* ScalarType in Executorch do not have support for below data types.
  * So, creating a placeholder for these data types. Once, ScalarTypes is
  * updated to have support for below data types, these can be removed and
@@ -294,7 +295,7 @@ Tensor& dequantize_impl(
       const auto* input_data_ptr = input.const_data_ptr<CTYPE_IN>();          \
       ET_CHECK_MSG(                                                           \
           *axis == 0, "Axis must be 0 for a single dimensional tensors");     \
-      const std::optional<int64_t> dim;                                       \
+      const optional<int64_t> dim;                                            \
       torch::executor::apply_over_dim(                                        \
           [input_data_ptr, out_data_ptr, zero_point_data, scale_data](        \
               size_t numel, size_t stride, size_t base_ix) {                  \
@@ -500,7 +501,7 @@ Tensor& dequantize_impl(
       const auto* input_data_ptr = input.const_data_ptr<CTYPE_IN>();          \
       ET_CHECK_MSG(                                                           \
           *axis == 0, "Axis must be 0 for a single dimensional tensors");     \
-      const std::optional<int64_t> dim;                                       \
+      const optional<int64_t> dim;                                            \
       torch::executor::apply_over_dim(                                        \
           [input_data_ptr, out_data_ptr, zero_point_data, scale_data](        \
               size_t numel, size_t stride, size_t base_ix) {                  \
diff --git a/extension/llm/custom_ops/op_sdpa.cpp b/extension/llm/custom_ops/op_sdpa.cpp
index 35a06011672..9ab0f99356a 100644
--- a/extension/llm/custom_ops/op_sdpa.cpp
+++ b/extension/llm/custom_ops/op_sdpa.cpp
@@ -38,7 +38,7 @@ bool validate_flash_attention_args(
     const Tensor& query,
     const Tensor& key,
     const Tensor& value,
-    const std::optional<Tensor>& attn_mask) {
+    const optional<Tensor>& attn_mask) {
   ET_CHECK_OR_RETURN_FALSE(query.dim() == 4, "query must be a 4D tensor");
   ET_CHECK_OR_RETURN_FALSE(key.dim() == 4, "key must be a 4D tensor");
   ET_CHECK_OR_RETURN_FALSE(value.dim() == 4, "value must be a 4D tensor");
@@ -325,11 +325,11 @@ Tensor& flash_attention_kernel_out(
     const Tensor& query,
     const Tensor& key,
     const Tensor& value,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output) {
   (void)ctx;
   ET_KERNEL_CHECK(
@@ -362,12 +362,12 @@ Tensor& flash_attention_kernel_out(
               is_causal,
               attn_mask,
               scale,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt);
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt);
         } else if (seq_len >= 192) {
           sdpa::impl::cpu_flash_attention<CTYPE, 64, 512>(
               ctx,
@@ -379,12 +379,12 @@ Tensor& flash_attention_kernel_out(
               is_causal,
               attn_mask,
               scale,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt);
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt);
         } else {
           sdpa::impl::cpu_flash_attention<CTYPE, 32, 512>(
               ctx,
@@ -396,12 +396,12 @@ Tensor& flash_attention_kernel_out(
               is_causal,
               attn_mask,
               scale,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt,
-              std::nullopt);
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt,
+              nullopt);
         }
       });
   return output;
@@ -413,18 +413,18 @@ Tensor& custom_sdpa_out_impl(
     const Tensor& k,
     const Tensor& v,
     const int64_t start_pos,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output,
-    const std::optional<Tensor>& q_zero_points = std::nullopt,
-    const std::optional<Tensor>& q_scales = std::nullopt,
-    const std::optional<Tensor>& k_zero_points = std::nullopt,
-    const std::optional<Tensor>& k_scales = std::nullopt,
-    const std::optional<Tensor>& v_zero_points = std::nullopt,
-    const std::optional<Tensor>& v_scales = std::nullopt,
+    const optional<Tensor>& q_zero_points = nullopt,
+    const optional<Tensor>& q_scales = nullopt,
+    const optional<Tensor>& k_zero_points = nullopt,
+    const optional<Tensor>& k_scales = nullopt,
+    const optional<Tensor>& v_zero_points = nullopt,
+    const optional<Tensor>& v_scales = nullopt,
     bool is_seq_at_dim_2 = false) {
   ET_KERNEL_CHECK_MSG(
       ctx,
@@ -570,17 +570,17 @@ Tensor& custom_quantized_sdpa_out(
     const Tensor& k,
     const Tensor& v,
     const int64_t start_pos,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
-    const std::optional<Tensor>& q_zero_points,
-    const std::optional<Tensor>& q_scales,
-    const std::optional<Tensor>& k_zero_points,
-    const std::optional<Tensor>& k_scales,
-    const std::optional<Tensor>& v_zero_points,
-    const std::optional<Tensor>& v_scales,
+    const optional<double> scale,
+    const optional<Tensor>& q_zero_points,
+    const optional<Tensor>& q_scales,
+    const optional<Tensor>& k_zero_points,
+    const optional<Tensor>& k_scales,
+    const optional<Tensor>& v_zero_points,
+    const optional<Tensor>& v_scales,
     const bool is_seq_at_dim_2,
     Tensor& output) {
   return custom_sdpa_out_impl(
@@ -624,11 +624,11 @@ Tensor& custom_sdpa_out(
     const Tensor& k,
     const Tensor& v,
     const int64_t start_pos,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output) {
   return custom_sdpa_out_impl(
       ctx, q, k, v, start_pos, attn_mask, dropout_p, is_causal, scale, output);
@@ -658,11 +658,11 @@ Tensor& sdpa_with_kv_cache_out(
     Tensor& value_cache,
     const int64_t start_pos,
     const int64_t seq_len,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output) {
   (void)ctx;
   ET_KERNEL_CHECK(
diff --git a/extension/llm/custom_ops/op_sdpa.h b/extension/llm/custom_ops/op_sdpa.h
index b416931467f..bb27a9badb8 100644
--- a/extension/llm/custom_ops/op_sdpa.h
+++ b/extension/llm/custom_ops/op_sdpa.h
@@ -26,11 +26,11 @@ Tensor& sdpa_with_kv_cache_out(
     Tensor& value_cache,
     const int64_t start_pos,
     const int64_t seq_len,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output);
 
 Tensor& custom_sdpa_out(
@@ -39,11 +39,11 @@ Tensor& custom_sdpa_out(
     const Tensor& k,
     const Tensor& v,
     const int64_t start_pos,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output);
 
 Tensor& flash_attention_kernel_out(
@@ -51,11 +51,11 @@ Tensor& flash_attention_kernel_out(
     const Tensor& query,
     const Tensor& key,
     const Tensor& value,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output);
 
 Tensor& custom_quantized_sdpa_out(
@@ -64,17 +64,17 @@ Tensor& custom_quantized_sdpa_out(
     const Tensor& k,
     const Tensor& v,
     const int64_t start_pos,
-    const std::optional<Tensor>& attn_mask,
+    const optional<Tensor>& attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
-    const std::optional<Tensor>& q_zero_points,
-    const std::optional<Tensor>& q_scales,
-    const std::optional<Tensor>& k_zero_points,
-    const std::optional<Tensor>& k_scales,
-    const std::optional<Tensor>& v_zero_points,
-    const std::optional<Tensor>& v_scales,
+    const optional<double> scale,
+    const optional<Tensor>& q_zero_points,
+    const optional<Tensor>& q_scales,
+    const optional<Tensor>& k_zero_points,
+    const optional<Tensor>& k_scales,
+    const optional<Tensor>& v_zero_points,
+    const optional<Tensor>& v_scales,
     const bool is_seq_at_dim_1,
     Tensor& output);
 
diff --git a/extension/llm/custom_ops/op_sdpa_aot.cpp b/extension/llm/custom_ops/op_sdpa_aot.cpp
index ba6f1275fd6..ce8fa88ce1b 100644
--- a/extension/llm/custom_ops/op_sdpa_aot.cpp
+++ b/extension/llm/custom_ops/op_sdpa_aot.cpp
@@ -27,11 +27,11 @@ Tensor& sdpa_with_kv_cache_out_no_context(
     const int64_t seq_len,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output);
 
 at::Tensor sdpa_with_kv_cache_aten(
@@ -57,11 +57,11 @@ Tensor& custom_sdpa_out_no_context(
     const int64_t start_pos,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output);
 
 at::Tensor custom_sdpa_aten(
@@ -84,17 +84,17 @@ Tensor& custom_quantized_sdpa_out_no_context(
     const int64_t start_pos,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
-    const std::optional<Tensor> q_zero_points,
-    const std::optional<Tensor> q_scales,
-    const std::optional<Tensor> k_zero_points,
-    const std::optional<Tensor> k_scales,
-    const std::optional<Tensor> v_zero_points,
-    const std::optional<Tensor> v_scales,
+    const optional<double> scale,
+    const optional<Tensor> q_zero_points,
+    const optional<Tensor> q_scales,
+    const optional<Tensor> k_zero_points,
+    const optional<Tensor> k_scales,
+    const optional<Tensor> v_zero_points,
+    const optional<Tensor> v_scales,
     const bool is_seq_at_dim_2,
     Tensor& output);
 
@@ -153,11 +153,11 @@ Tensor& sdpa_with_kv_cache_out_no_context(
     const int64_t seq_len,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output) {
   executorch::runtime::KernelRuntimeContext context{};
   return torch::executor::native::sdpa_with_kv_cache_out(
@@ -215,11 +215,11 @@ Tensor& custom_sdpa_out_no_context(
     const int64_t start_pos,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
+    const optional<double> scale,
     Tensor& output) {
   executorch::aten::RuntimeContext context{};
   return torch::executor::native::custom_sdpa_out(
@@ -260,17 +260,17 @@ Tensor& custom_quantized_sdpa_out_no_context(
     const int64_t start_pos,
     // @lint-ignore CLANGTIDY facebook-hte-ConstantArgumentPassByValue
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<Tensor> attn_mask,
+    const optional<Tensor> attn_mask,
     const double dropout_p,
     const bool is_causal,
     // @lint-ignore CLANGTIDY facebook-hte-ParameterMightThrowOnCopy
-    const std::optional<double> scale,
-    const std::optional<Tensor> q_zero_points,
-    const std::optional<Tensor> q_scales,
-    const std::optional<Tensor> k_zero_points,
-    const std::optional<Tensor> k_scales,
-    const std::optional<Tensor> v_zero_points,
-    const std::optional<Tensor> v_scales,
+    const optional<double> scale,
+    const optional<Tensor> q_zero_points,
+    const optional<Tensor> q_scales,
+    const optional<Tensor> k_zero_points,
+    const optional<Tensor> k_scales,
+    const optional<Tensor> v_zero_points,
+    const optional<Tensor> v_scales,
     const bool is_seq_at_dim_2,
     Tensor& output) {
   executorch::aten::RuntimeContext context{};
diff --git a/extension/llm/custom_ops/op_sdpa_impl.h b/extension/llm/custom_ops/op_sdpa_impl.h
index 90750bfc114..9c479569505 100644
--- a/extension/llm/custom_ops/op_sdpa_impl.h
+++ b/extension/llm/custom_ops/op_sdpa_impl.h
@@ -613,14 +613,14 @@ void cpu_flash_attention(
     const Tensor& value,
     double dropout_p,
     bool is_causal,
-    const std::optional<Tensor>& attn_mask,
-    const std::optional<double>& scale,
-    const std::optional<Tensor>& q_zero_points,
-    const std::optional<Tensor>& q_scales,
-    const std::optional<Tensor>& k_zero_points,
-    const std::optional<Tensor>& k_scales,
-    const std::optional<Tensor>& v_zero_points,
-    const std::optional<Tensor>& v_scales,
+    const optional<Tensor>& attn_mask,
+    const optional<double>& scale,
+    const optional<Tensor>& q_zero_points,
+    const optional<Tensor>& q_scales,
+    const optional<Tensor>& k_zero_points,
+    const optional<Tensor>& k_scales,
+    const optional<Tensor>& v_zero_points,
+    const optional<Tensor>& v_scales,
     const SeqDim seq_dim = SeqDim::TWO,
     const int64_t start_pos = 0,
     const int64_t num_keys_for_causal_attention = -1) {
diff --git a/extension/llm/custom_ops/op_update_cache.cpp b/extension/llm/custom_ops/op_update_cache.cpp
index 215cb276a28..7ab994deb5f 100644
--- a/extension/llm/custom_ops/op_update_cache.cpp
+++ b/extension/llm/custom_ops/op_update_cache.cpp
@@ -26,7 +26,7 @@ bool validate_cache_params(
     const Tensor& quantized_cache,
     int64_t start_pos,
     int64_t seq_length,
-    const std::optional<Tensor>& indices = std::nullopt) {
+    const optional<Tensor>& indices = nullopt) {
   ET_CHECK_OR_RETURN_FALSE(
       quantized_cache.dim() == 4, "quantized cache must be a 4D tensor");
 
@@ -94,7 +94,7 @@ Tensor& update_cache_impl(
     Tensor& cache,
     const int64_t start_pos,
     Tensor& output,
-    const std::optional<Tensor>& indices = std::nullopt) {
+    const optional<Tensor>& indices = nullopt) {
   (void)ctx;
 
   ET_CHECK_MSG(
diff --git a/kernels/optimized/cpu/op_gelu.cpp b/kernels/optimized/cpu/op_gelu.cpp
index 2649c50424b..a36d3c259c6 100644
--- a/kernels/optimized/cpu/op_gelu.cpp
+++ b/kernels/optimized/cpu/op_gelu.cpp
@@ -23,6 +23,7 @@ namespace native {
 
 using Tensor = executorch::aten::Tensor;
 using ScalarType = executorch::aten::ScalarType;
+using string_view = std::string_view;
 
 namespace {
 
@@ -40,7 +41,7 @@ template <typename CTYPE>
 void gelu(
     executorch::runtime::KernelRuntimeContext& context,
     const Tensor& input,
-    std::string_view approximate,
+    string_view approximate,
     Tensor& output) {
   const CTYPE* in_data = input.const_data_ptr<CTYPE>();
   CTYPE* out_data = output.mutable_data_ptr<CTYPE>();
@@ -90,7 +91,7 @@ void gelu(
 Tensor& opt_gelu_out(
     KernelRuntimeContext& context,
     const Tensor& input,
-    std::string_view approximate,
+    string_view approximate,
     Tensor& out) {
   (void)context;
   ET_KERNEL_CHECK(
diff --git a/kernels/optimized/cpu/op_linear.cpp b/kernels/optimized/cpu/op_linear.cpp
index 7fbe5624651..382e6e19415 100644
--- a/kernels/optimized/cpu/op_linear.cpp
+++ b/kernels/optimized/cpu/op_linear.cpp
@@ -73,7 +73,7 @@ Tensor& opt_linear_out(
     RuntimeContext& ctx,
     const Tensor& in,
     const Tensor& mat2,
-    const std::optional<Tensor>& bias,
+    const optional<Tensor>& bias,
     Tensor& out) {
   ET_KERNEL_CHECK(ctx, check_linear_args(in, mat2, out), InvalidArgument, out);
 
diff --git a/kernels/optimized/cpu/op_native_layer_norm.cpp b/kernels/optimized/cpu/op_native_layer_norm.cpp
index eecf1317610..5fac9faf25e 100644
--- a/kernels/optimized/cpu/op_native_layer_norm.cpp
+++ b/kernels/optimized/cpu/op_native_layer_norm.cpp
@@ -27,8 +27,8 @@ template <typename CTYPE>
 void layer_norm(
     const Tensor& input,
     IntArrayRef normalized_shape,
-    const std::optional<Tensor>& weight,
-    const std::optional<Tensor>& bias,
+    const optional<Tensor>& weight,
+    const optional<Tensor>& bias,
     CTYPE eps,
     Tensor& out,
     Tensor& mean,
diff --git a/kernels/portable/cpu/op_any.cpp b/kernels/portable/cpu/op_any.cpp
index 1acf484d3f0..0f3a36b6ba7 100644
--- a/kernels/portable/cpu/op_any.cpp
+++ b/kernels/portable/cpu/op_any.cpp
@@ -54,7 +54,7 @@ Tensor& any_all_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
 Tensor& any_dims_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool keepdim,
     Tensor& out) {
   (void)ctx;
diff --git a/kernels/portable/cpu/op_as_strided_copy.cpp b/kernels/portable/cpu/op_as_strided_copy.cpp
index 556813b5fef..060e3cf0761 100644
--- a/kernels/portable/cpu/op_as_strided_copy.cpp
+++ b/kernels/portable/cpu/op_as_strided_copy.cpp
@@ -21,7 +21,7 @@ Tensor& as_strided_copy_out(
     const Tensor& in,
     ArrayRef<int64_t> size,
     ArrayRef<int64_t> stride,
-    std::optional<int64_t> storage_offset,
+    optional<int64_t> storage_offset,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_cumsum.cpp b/kernels/portable/cpu/op_cumsum.cpp
index e6374d7d66d..3a518d30715 100644
--- a/kernels/portable/cpu/op_cumsum.cpp
+++ b/kernels/portable/cpu/op_cumsum.cpp
@@ -90,7 +90,7 @@ Tensor& cumsum_out(
     KernelRuntimeContext& ctx,
     const Tensor& self,
     int64_t dim,
-    std::optional<ScalarType> enforced_dtype,
+    optional<ScalarType> enforced_dtype,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_full_like.cpp b/kernels/portable/cpu/op_full_like.cpp
index 32b4d720d53..5fefd53c30b 100644
--- a/kernels/portable/cpu/op_full_like.cpp
+++ b/kernels/portable/cpu/op_full_like.cpp
@@ -21,7 +21,7 @@ Tensor& full_like_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
     const Scalar& fill_value,
-    std::optional<MemoryFormat> memory_format,
+    optional<MemoryFormat> memory_format,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_gelu.cpp b/kernels/portable/cpu/op_gelu.cpp
index 8daec086dde..0489d3d12b4 100644
--- a/kernels/portable/cpu/op_gelu.cpp
+++ b/kernels/portable/cpu/op_gelu.cpp
@@ -19,11 +19,12 @@ namespace native {
 
 using Tensor = executorch::aten::Tensor;
 using ScalarType = executorch::aten::ScalarType;
+using string_view = std::string_view;
 
 Tensor& gelu_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::string_view approximate,
+    string_view approximate,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_mean.cpp b/kernels/portable/cpu/op_mean.cpp
index bd8733e4229..9d905753250 100644
--- a/kernels/portable/cpu/op_mean.cpp
+++ b/kernels/portable/cpu/op_mean.cpp
@@ -24,9 +24,9 @@ using ScalarType = executorch::aten::ScalarType;
 Tensor& mean_dim_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   (void)ctx;
 
@@ -123,7 +123,7 @@ Tensor& mean_dim_out(
 Tensor& mean_dtype_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   return mean_dim_out(ctx, in, ArrayRef<int64_t>(), false, dtype, out);
 }
diff --git a/kernels/portable/cpu/op_native_group_norm.cpp b/kernels/portable/cpu/op_native_group_norm.cpp
index 7ec411f1355..7a2b40cba1a 100644
--- a/kernels/portable/cpu/op_native_group_norm.cpp
+++ b/kernels/portable/cpu/op_native_group_norm.cpp
@@ -24,8 +24,8 @@ namespace {
 template <typename CTYPE>
 void group_norm(
     const Tensor& input,
-    const std::optional<Tensor>& weight,
-    const std::optional<Tensor>& bias,
+    const optional<Tensor>& weight,
+    const optional<Tensor>& bias,
     int64_t sN,
     int64_t sC,
     int64_t sHxW,
diff --git a/kernels/portable/cpu/op_native_layer_norm.cpp b/kernels/portable/cpu/op_native_layer_norm.cpp
index 1e0ee3b82e0..b35796f3328 100644
--- a/kernels/portable/cpu/op_native_layer_norm.cpp
+++ b/kernels/portable/cpu/op_native_layer_norm.cpp
@@ -24,8 +24,8 @@ template <typename CTYPE>
 void layer_norm(
     const Tensor& input,
     IntArrayRef normalized_shape,
-    const std::optional<Tensor>& weight,
-    const std::optional<Tensor>& bias,
+    const optional<Tensor>& weight,
+    const optional<Tensor>& bias,
     CTYPE eps,
     Tensor& out,
     Tensor& mean,
diff --git a/kernels/portable/cpu/op_prod.cpp b/kernels/portable/cpu/op_prod.cpp
index e195d3c6504..ba76a1f200c 100644
--- a/kernels/portable/cpu/op_prod.cpp
+++ b/kernels/portable/cpu/op_prod.cpp
@@ -20,7 +20,7 @@ using ScalarType = executorch::aten::ScalarType;
 Tensor& prod_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   (void)ctx;
 
@@ -53,7 +53,7 @@ Tensor& prod_int_out(
     const Tensor& in,
     int64_t dim,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_sum.cpp b/kernels/portable/cpu/op_sum.cpp
index a6a1c417933..194d909a9e3 100644
--- a/kernels/portable/cpu/op_sum.cpp
+++ b/kernels/portable/cpu/op_sum.cpp
@@ -25,9 +25,9 @@ using ScalarType = executorch::aten::ScalarType;
 Tensor& sum_dim_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   (void)ctx;
 
diff --git a/kernels/portable/cpu/op_var.cpp b/kernels/portable/cpu/op_var.cpp
index 3901fdd8376..202d7df80bc 100644
--- a/kernels/portable/cpu/op_var.cpp
+++ b/kernels/portable/cpu/op_var.cpp
@@ -24,7 +24,7 @@ void compute_variance(
     KernelRuntimeContext& ctx,
     const Tensor& in,
     Tensor& out,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     const size_t num,
     const double denominator) {
   CTYPE_OUT* out_data = out.mutable_data_ptr<CTYPE_OUT>();
@@ -62,7 +62,7 @@ void compute_variance(
 Tensor& var_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool unbiased,
     bool keepdim,
     Tensor& out) {
@@ -105,8 +105,8 @@ Tensor& var_out(
 Tensor& var_correction_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
-    const std::optional<Scalar>& correction,
+    optional<ArrayRef<int64_t>> dim_list,
+    const optional<Scalar>& correction,
     bool keepdim,
     Tensor& out) {
   (void)ctx;
diff --git a/kernels/portable/cpu/op_var_mean.cpp b/kernels/portable/cpu/op_var_mean.cpp
index 9f285f8f4b0..dbb88f2f32d 100644
--- a/kernels/portable/cpu/op_var_mean.cpp
+++ b/kernels/portable/cpu/op_var_mean.cpp
@@ -24,7 +24,7 @@ void compute_var_mean(
     const Tensor& in,
     Tensor& var_out,
     Tensor& mean_out,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     const size_t num,
     const double denominator) {
   CTYPE_OUT* var_data = var_out.mutable_data_ptr<CTYPE_OUT>();
@@ -102,8 +102,8 @@ void compute_var_mean(
 std::tuple<Tensor&, Tensor&> var_mean_correction_out(
     KernelRuntimeContext& ctx,
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
-    const std::optional<Scalar>& correction,
+    optional<ArrayRef<int64_t>> dim_list,
+    const optional<Scalar>& correction,
     bool keepdim,
     Tensor& out0,
     Tensor& out1) {
diff --git a/kernels/portable/cpu/util/copy_ops_util.cpp b/kernels/portable/cpu/util/copy_ops_util.cpp
index b1eb1760fbd..16c0a32c62b 100644
--- a/kernels/portable/cpu/util/copy_ops_util.cpp
+++ b/kernels/portable/cpu/util/copy_ops_util.cpp
@@ -43,7 +43,7 @@ bool check_as_strided_copy_args(
     const Tensor& in,
     ArrayRef<int64_t> size,
     ArrayRef<int64_t> stride,
-    std::optional<int64_t> storage_offset,
+    optional<int64_t> storage_offset,
     Tensor& out) {
   ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, out));
   ET_CHECK_OR_RETURN_FALSE(
diff --git a/kernels/portable/cpu/util/copy_ops_util.h b/kernels/portable/cpu/util/copy_ops_util.h
index 68f3309b044..af2e8c83cad 100644
--- a/kernels/portable/cpu/util/copy_ops_util.h
+++ b/kernels/portable/cpu/util/copy_ops_util.h
@@ -58,7 +58,7 @@ bool check_as_strided_copy_args(
     const Tensor& in,
     ArrayRef<int64_t> size,
     ArrayRef<int64_t> stride,
-    std::optional<int64_t> storage_offset,
+    optional<int64_t> storage_offset,
     Tensor& out);
 
 template <typename CTYPE>
diff --git a/kernels/portable/cpu/util/distance_util.cpp b/kernels/portable/cpu/util/distance_util.cpp
index a3cf4cf318e..e7f146e2e9d 100644
--- a/kernels/portable/cpu/util/distance_util.cpp
+++ b/kernels/portable/cpu/util/distance_util.cpp
@@ -32,7 +32,7 @@ bool check_cdist_args(
     const Tensor& x1,
     const Tensor& x2,
     double p,
-    std::optional<int64_t> compute_mode,
+    optional<int64_t> compute_mode,
     const Tensor& out) {
   ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(x1, x2));
   ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(x1, out));
diff --git a/kernels/portable/cpu/util/distance_util.h b/kernels/portable/cpu/util/distance_util.h
index 32f9d91a290..a54e02dab28 100644
--- a/kernels/portable/cpu/util/distance_util.h
+++ b/kernels/portable/cpu/util/distance_util.h
@@ -127,7 +127,7 @@ bool check_cdist_args(
     const Tensor& x1,
     const Tensor& x2,
     double p,
-    std::optional<int64_t> compute_mode,
+    optional<int64_t> compute_mode,
     const Tensor& out);
 
 } // namespace executor
diff --git a/kernels/portable/cpu/util/kernel_ops_util.cpp b/kernels/portable/cpu/util/kernel_ops_util.cpp
index 484d7e74fd4..2e0ceee07e5 100644
--- a/kernels/portable/cpu/util/kernel_ops_util.cpp
+++ b/kernels/portable/cpu/util/kernel_ops_util.cpp
@@ -507,7 +507,7 @@ void get_convolution_out_target_size(
 bool check_cumsum_args(
     const Tensor& in,
     int64_t dim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   ET_LOG_AND_RETURN_IF_FALSE(dim_is_valid(dim, in.dim()));
 
diff --git a/kernels/portable/cpu/util/kernel_ops_util.h b/kernels/portable/cpu/util/kernel_ops_util.h
index a6f95d0899e..db6ce070542 100644
--- a/kernels/portable/cpu/util/kernel_ops_util.h
+++ b/kernels/portable/cpu/util/kernel_ops_util.h
@@ -445,7 +445,7 @@ void get_convolution_out_target_size(
 bool check_cumsum_args(
     const Tensor& self,
     int64_t dim,
-    std::optional<ScalarType> enforced_dtype,
+    optional<ScalarType> enforced_dtype,
     Tensor& out);
 
 bool check_max_pool2d_with_indices_args(
diff --git a/kernels/portable/cpu/util/reduce_util.cpp b/kernels/portable/cpu/util/reduce_util.cpp
index d622e758a4e..afeb56f719f 100644
--- a/kernels/portable/cpu/util/reduce_util.cpp
+++ b/kernels/portable/cpu/util/reduce_util.cpp
@@ -319,9 +319,9 @@ Error resize_reduction_out(
  */
 bool check_reduction_args(
     const Tensor& in,
-    const std::optional<ArrayRef<int64_t>>& dim_list,
+    const optional<ArrayRef<int64_t>>& dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   if (dtype.has_value()) {
     ET_LOG_AND_RETURN_IF_FALSE(dtype.value() == out.scalar_type());
@@ -339,9 +339,9 @@ bool check_reduction_args(
  */
 bool check_reduction_args_single_dim(
     const Tensor& in,
-    std::optional<int64_t> dim,
+    optional<int64_t> dim,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out,
     bool allow_empty_dim) {
   if (dtype.has_value()) {
@@ -369,9 +369,9 @@ bool check_reduction_args_single_dim(
 
 bool check_mean_dim_args(
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   ET_LOG_AND_RETURN_IF_FALSE(
       check_reduction_args(in, dim_list, keepdim, dtype, out));
@@ -402,7 +402,7 @@ bool check_amin_amax_args(
 
 bool check_argmin_argmax_args(
     const Tensor& in,
-    std::optional<int64_t> dim,
+    optional<int64_t> dim,
     bool keepdim,
     Tensor& out) {
   ET_LOG_AND_RETURN_IF_FALSE(
@@ -432,7 +432,7 @@ bool check_min_max_args(
 
 bool check_prod_out_args(
     const Tensor& in,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out) {
   if (dtype.has_value()) {
     ET_LOG_AND_RETURN_IF_FALSE(dtype.value() == out.scalar_type());
diff --git a/kernels/portable/cpu/util/reduce_util.h b/kernels/portable/cpu/util/reduce_util.h
index bf61b0d9090..f800fb6f736 100644
--- a/kernels/portable/cpu/util/reduce_util.h
+++ b/kernels/portable/cpu/util/reduce_util.h
@@ -755,24 +755,24 @@ inline Error resize_reduction_out(
 #ifndef USE_ATEN_LIB
 bool check_reduction_args(
     const Tensor& in,
-    const std::optional<ArrayRef<int64_t>>& dim_list,
+    const optional<ArrayRef<int64_t>>& dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out);
 
 bool check_reduction_args_single_dim(
     const Tensor& in,
-    std::optional<int64_t> dim,
+    optional<int64_t> dim,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out,
     bool allow_empty_dim = false);
 
 bool check_mean_dim_args(
     const Tensor& in,
-    std::optional<ArrayRef<int64_t>> dim_list,
+    optional<ArrayRef<int64_t>> dim_list,
     bool keepdim,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out);
 
 bool check_amin_amax_args(
@@ -783,7 +783,7 @@ bool check_amin_amax_args(
 
 bool check_argmin_argmax_args(
     const Tensor& in,
-    std::optional<int64_t> dim,
+    optional<int64_t> dim,
     bool keepdim,
     Tensor& out);
 
@@ -796,7 +796,7 @@ bool check_min_max_args(
 
 bool check_prod_out_args(
     const Tensor& in,
-    std::optional<ScalarType> dtype,
+    optional<ScalarType> dtype,
     Tensor& out);
 
 #endif
diff --git a/runtime/core/exec_aten/exec_aten.h b/runtime/core/exec_aten/exec_aten.h
index d3c447df8db..ac4fb9a126e 100644
--- a/runtime/core/exec_aten/exec_aten.h
+++ b/runtime/core/exec_aten/exec_aten.h
@@ -14,6 +14,7 @@
 #include <executorch/runtime/core/error.h> // @manual
 #include <executorch/runtime/core/result.h> // @manual
 #include <executorch/runtime/core/tensor_shape_dynamism.h> // @manual
+#include <executorch/runtime/platform/assert.h> // @manual
 #include <executorch/runtime/platform/compiler.h>
 #ifdef USE_ATEN_LIB
 #include <ATen/Tensor.h> // @manual
@@ -26,6 +27,7 @@
 #include <c10/util/BFloat16-math.h> // @manual
 #include <c10/util/BFloat16.h> // @manual
 #include <c10/util/Half.h> // @manual
+#include <c10/util/Optional.h> // @manual
 #include <c10/util/complex.h> // @manual
 #include <c10/util/qint32.h> // @manual
 #include <c10/util/qint8.h> // @manual
@@ -33,6 +35,7 @@
 #include <c10/util/quint4x2.h> // @manual
 #include <c10/util/quint8.h> // @manual
 #include <c10/util/safe_numerics.h> // @manual
+#include <c10/util/string_view.h> // @manual
 #include <torch/torch.h>
 #else // use executor
 #include <executorch/runtime/core/array_ref.h> // @manual
diff --git a/runtime/core/exec_aten/util/scalar_type_util.h b/runtime/core/exec_aten/util/scalar_type_util.h
index 7a18d94bacc..3e8e36b442e 100644
--- a/runtime/core/exec_aten/util/scalar_type_util.h
+++ b/runtime/core/exec_aten/util/scalar_type_util.h
@@ -25,7 +25,6 @@
 #include <cinttypes>
 #include <cstdint>
 #include <limits>
-#include <string_view>
 #include <type_traits>
 
 #include <c10/macros/Macros.h>
@@ -48,10 +47,11 @@ using ScalarType = at::ScalarType;
 } // namespace executorch
 #else // !USE_ATEN_LIB
 #include <executorch/runtime/core/portable_type/scalar_type.h>
+#include <executorch/runtime/core/portable_type/string_view.h>
 namespace executorch {
 namespace aten {
 using ScalarType = torch::executor::ScalarType;
-using string_view ET_DEPRECATED = std::string_view;
+using string_view = std::string_view;
 } // namespace aten
 } // namespace executorch
 #endif // USE_ATEN_LIB