diff --git a/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_deserialize.cpp b/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_deserialize.cpp
index 26515a1b304..2b0e2f6785b 100644
--- a/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_deserialize.cpp
+++ b/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_deserialize.cpp
@@ -106,7 +106,7 @@ c10::intrusive_ptr<LinearPackedParamsBase> PackedLinearWeight::deserialize(
         std::get<weight_scales_index>(serialized),
         weight_zero_points,
         0, // The output channel axis is 0
-        device(c10::kCPU).dtype(c10::kQInt8));
+        at::device(c10::kCPU).dtype(c10::kQInt8));
   }
 
   const at::Tensor loaded_weight_values =
diff --git a/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_unpack.cpp b/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_unpack.cpp
index 196e711da7e..bda1984d620 100644
--- a/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_unpack.cpp
+++ b/aten/src/ATen/native/ao_sparse/quantized/cpu/qlinear_unpack.cpp
@@ -46,7 +46,7 @@ LinearPackedSerializationType PackedLinearWeight::unpack() {
         scales,
         zero_points,
         0, // The output channel axis is 0
-        device(c10::kCPU).dtype(c10::kQInt8));
+        at::device(c10::kCPU).dtype(c10::kQInt8));
   }
 
   int8_t* weight_ptr_int8 =
@@ -100,7 +100,7 @@ LinearPackedSerializationType PackedLinearWeightQnnp::unpack() {
         scales,
         zero_points,
         0, // The output channel axis is 0
-        device(c10::kCPU).dtype(c10::kQInt8));
+        at::device(c10::kCPU).dtype(c10::kQInt8));
   }
 
   int8_t* weight_ptr_int8 =
diff --git a/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp b/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
index c9c09cf2464..f62db511b36 100644
--- a/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
+++ b/aten/src/ATen/native/quantized/cpu/LinearUnpackImpl.cpp
@@ -32,16 +32,16 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedLinearWeight::unpack() {
         {N, K}, at::device(c10::kCPU).dtype(c10::kQInt8), w_scale[0], w_zp[0]);
   } else if (q_scheme == c10::kPerChannelAffine) {
     auto scales = at::from_blob(
-        w_scale.data(), w_scale.size(), device(c10::kCPU).dtype(c10::kFloat));
+        w_scale.data(), w_scale.size(), at::device(c10::kCPU).dtype(c10::kFloat));
     auto zero_points = at::from_blob(
-        w_zp.data(), w_zp.size(), device(c10::kCPU).dtype(c10::kInt));
+        w_zp.data(), w_zp.size(), at::device(c10::kCPU).dtype(c10::kInt));
 
     weight_origin = at::_empty_per_channel_affine_quantized(
         {N, K},
         scales.toType(c10::kDouble),
         zero_points.toType(c10::kLong),
         0, // The output channel axis is 0
-        device(c10::kCPU).dtype(c10::kQInt8));
+        at::device(c10::kCPU).dtype(c10::kQInt8));
   }
 
   int8_t* weight_ptr_int8 =
@@ -81,7 +81,7 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedLinearWeightsQnnp::
       auto scales = at::from_blob(
           weight_scales_data,
           w_scales.sizes()[0] - kPaddingChannels,
-          device(c10::kCPU).dtype(c10::kFloat));
+          at::device(c10::kCPU).dtype(c10::kFloat));
 
       at::Tensor zero_points = at::empty(
           w_zero_points.size() - kPaddingChannels, at::device(c10::kCPU).dtype(c10::kLong));
@@ -93,7 +93,7 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedLinearWeightsQnnp::
                           scales,
                           zero_points.toType(c10::kLong),
                           0, // The output channel axis is 0
-                          device(c10::kCPU).dtype(c10::kQInt8))
+                          at::device(c10::kCPU).dtype(c10::kQInt8))
                           .contiguous();
     } else {
       TORCH_INTERNAL_ASSERT(false, "Unsupported quantization scheme.");
diff --git a/aten/src/ATen/native/quantized/cpu/qconv.cpp b/aten/src/ATen/native/quantized/cpu/qconv.cpp
index 9acdd0937c8..98e9a5af8a9 100644
--- a/aten/src/ATen/native/quantized/cpu/qconv.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qconv.cpp
@@ -448,7 +448,7 @@ at::Tensor PackedConvWeight<kSpatialDim>::apply_impl(
   at::Tensor output = kSpatialDim == 2
       ? at::_empty_affine_quantized(
             output_shape,
-            device(c10::kCPU)
+            at::device(c10::kCPU)
                 .dtype(c10::kQUInt8)
                 .memory_format(c10::MemoryFormat::ChannelsLast),
             output_scale,
@@ -460,7 +460,7 @@ at::Tensor PackedConvWeight<kSpatialDim>::apply_impl(
             output_shape[2],
             output_shape[3],
             output_shape[4],
-            device(c10::kCPU).dtype(c10::kQUInt8),
+            at::device(c10::kCPU).dtype(c10::kQUInt8),
             output_scale,
             output_zero_point);
   at::Tensor buffer =
@@ -1225,7 +1225,7 @@ at::Tensor PackedConvWeightsOnednn<kSpatialDim>::apply_impl(
   ideep::dims dst_dims = ideep::dims({output_sizes.cbegin(), output_sizes.cend()});
   at::Tensor output = at::_empty_affine_quantized(
       dst_dims,
-      device(c10::kCPU)
+      at::device(c10::kCPU)
           .dtype(c10::kQUInt8)
           .memory_format(kSpatialDim == 2 ?
               c10::MemoryFormat::ChannelsLast :
@@ -1593,7 +1593,7 @@ static at::Tensor _quantized_convolution_onednn(
     accum.value() :
     at::empty(
       dst_dims,
-      device(c10::kCPU)
+      at::device(c10::kCPU)
           .dtype(fp32_output ? c10::kFloat : (bfloat16_output ? c10::kBFloat16 : c10::kByte))
           .memory_format(kSpatialDim == 2 ?
               c10::MemoryFormat::ChannelsLast :
diff --git a/aten/src/ATen/native/quantized/cpu/qconv_unpack_impl.cpp b/aten/src/ATen/native/quantized/cpu/qconv_unpack_impl.cpp
index 83273f979e9..9ade8d3f212 100644
--- a/aten/src/ATen/native/quantized/cpu/qconv_unpack_impl.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qconv_unpack_impl.cpp
@@ -37,7 +37,7 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedConvWeight<
     unpacked_weights = kSpatialDim == 2
         ? at::_empty_affine_quantized(
               {output_channels, C_per_G, kernel_h, kernel_w},
-              device(c10::kCPU)
+              at::device(c10::kCPU)
                   .dtype(c10::kQInt8)
                   .memory_format(c10::MemoryFormat::ChannelsLast),
               w_scale[0],
@@ -50,7 +50,7 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedConvWeight<
                   kernel_d,
                   kernel_h,
                   kernel_w,
-                  device(c10::kCPU).dtype(c10::kQInt8),
+                  at::device(c10::kCPU).dtype(c10::kQInt8),
                   w_scale[0],
                   w_zp[0]);
   } else if (q_scheme == c10::kPerChannelAffine) {
@@ -58,16 +58,16 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedConvWeight<
         !transpose(),
         "Per Channel Quantization is currently disabled for transposed conv");
     auto scales = at::from_blob(
-        w_scale.data(), w_scale.size(), device(c10::kCPU).dtype(c10::kFloat));
+        w_scale.data(), w_scale.size(), at::device(c10::kCPU).dtype(c10::kFloat));
     auto zero_points = at::from_blob(
-        w_zp.data(), w_zp.size(), device(c10::kCPU).dtype(c10::kInt));
+        w_zp.data(), w_zp.size(), at::device(c10::kCPU).dtype(c10::kInt));
     unpacked_weights = kSpatialDim == 2
         ? at::_empty_per_channel_affine_quantized(
               {output_channels, C_per_G, kernel_h, kernel_w},
               scales.toType(c10::kDouble),
               zero_points.toType(c10::kLong),
               0, /* The output channel axis is 0 */
-              device(c10::kCPU).dtype(c10::kQInt8),
+              at::device(c10::kCPU).dtype(c10::kQInt8),
               c10::MemoryFormat::ChannelsLast)
         : at::native::fbgemm_utils::
               MakeEmptyPerChannelAffineQuantizedChannelsLast3dTensor(
@@ -76,7 +76,7 @@ std::tuple<at::Tensor, std::optional<at::Tensor>> PackedConvWeight<
                   kernel_d,
                   kernel_h,
                   kernel_w,
-                  device(c10::kCPU).dtype(c10::kQInt8),
+                  at::device(c10::kCPU).dtype(c10::kQInt8),
                   scales.toType(c10::kDouble),
                   zero_points.toType(c10::kLong));
   } else {
diff --git a/aten/src/ATen/native/quantized/cpu/qembeddingbag_unpack.cpp b/aten/src/ATen/native/quantized/cpu/qembeddingbag_unpack.cpp
index 7c023ffd56a..10e152aa0f1 100644
--- a/aten/src/ATen/native/quantized/cpu/qembeddingbag_unpack.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qembeddingbag_unpack.cpp
@@ -44,9 +44,9 @@ at::Tensor PackedEmbeddingBagWeight::unpack() {
             num_elem_per_byte};
 
     auto scales = at::from_blob(
-        w_scale.data(), w_scale.size(), device(c10::kCPU).dtype(c10::kFloat));
+        w_scale.data(), w_scale.size(), at::device(c10::kCPU).dtype(c10::kFloat));
     auto zero_points = at::from_blob(
-        w_zp.data(), w_zp.size(), device(c10::kCPU).dtype(c10::kFloat));
+        w_zp.data(), w_zp.size(), at::device(c10::kCPU).dtype(c10::kFloat));
 
     auto output_columns = output_shape[1];
     uint8_t* output_data = nullptr;
@@ -58,7 +58,7 @@ at::Tensor PackedEmbeddingBagWeight::unpack() {
           scales.toType(c10::kFloat),
           zero_points.toType(c10::kFloat),
           0, // The output channel axis is 0
-          device(c10::kCPU).dtype(c10::kQUInt8));
+          at::device(c10::kCPU).dtype(c10::kQUInt8));
       output_data = static_cast<uint8_t*>(weight_origin.data_ptr());
     } else {
       // We create empty qtensor with the full output shape, and dtype set to
@@ -69,7 +69,7 @@ at::Tensor PackedEmbeddingBagWeight::unpack() {
           scales.toType(c10::kFloat),
           zero_points.toType(c10::kFloat),
           0, // The output channel axis is 0
-          device(c10::kCPU).dtype(c10::kQUInt4x2));
+          at::device(c10::kCPU).dtype(c10::kQUInt4x2));
       output_data = static_cast<uint8_t*>(weight_origin.data_ptr());
     }
 
diff --git a/aten/src/ATen/native/quantized/cpu/qlinear.cpp b/aten/src/ATen/native/quantized/cpu/qlinear.cpp
index f1a81269a44..10f34a685f3 100644
--- a/aten/src/ATen/native/quantized/cpu/qlinear.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qlinear.cpp
@@ -1015,7 +1015,7 @@ static at::Tensor linear_int8_with_onednn_weight(
       other.value() :
       at::empty(
         dst_dims,
-        device(c10::kCPU)
+        at::device(c10::kCPU)
             .dtype(fp32_output ? c10::kFloat : (bf16_output ? c10::kBFloat16 : c10::kByte))
       );
   if (output.numel() == 0) {
diff --git a/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp b/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
index 5db6a6e14c4..e2d5278d579 100644
--- a/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
+++ b/aten/src/ATen/native/quantized/cpu/qlinear_dynamic.cpp
@@ -652,7 +652,7 @@ static at::Tensor linear_dynamic_fp16_with_onednn_weight(
   std::vector<int64_t> dst_dims = {M, N};
   at::Tensor output = at::empty(
         dst_dims,
-        device(c10::kCPU)
+        at::device(c10::kCPU)
             .dtype(c10::kFloat)
       );
   if (output.numel() == 0) {