diff --git a/torchvision/csrc/models/densenet.cpp b/torchvision/csrc/models/densenet.cpp
index 0e3a46e04ff..145748b1449 100644
--- a/torchvision/csrc/models/densenet.cpp
+++ b/torchvision/csrc/models/densenet.cpp
@@ -76,7 +76,7 @@ struct _TransitionImpl : torch::nn::SequentialImpl {
         torch::nn::Conv2d(Options(num_input_features, num_output_features, 1)
                               .stride(1)
                               .bias(false)));
-    push_back("pool", torch::nn::Functional([](torch::Tensor input) {
+    push_back("pool", torch::nn::Functional([](const torch::Tensor& input) {
                 return torch::avg_pool2d(input, 2, 2, 0, false, true);
               }));
   }
@@ -91,7 +91,7 @@ TORCH_MODULE(_Transition);
 DenseNetImpl::DenseNetImpl(
     int64_t num_classes,
     int64_t growth_rate,
-    std::vector<int64_t> block_config,
+    const std::vector<int64_t>& block_config,
     int64_t num_init_features,
     int64_t bn_size,
     double drop_rate) {
@@ -157,7 +157,7 @@ torch::Tensor DenseNetImpl::forward(torch::Tensor x) {
 DenseNet121Impl::DenseNet121Impl(
     int64_t num_classes,
     int64_t growth_rate,
-    std::vector<int64_t> block_config,
+    const std::vector<int64_t>& block_config,
     int64_t num_init_features,
     int64_t bn_size,
     double drop_rate)
@@ -172,7 +172,7 @@ DenseNet121Impl::DenseNet121Impl(
 DenseNet169Impl::DenseNet169Impl(
     int64_t num_classes,
     int64_t growth_rate,
-    std::vector<int64_t> block_config,
+    const std::vector<int64_t>& block_config,
     int64_t num_init_features,
     int64_t bn_size,
     double drop_rate)
@@ -187,7 +187,7 @@ DenseNet169Impl::DenseNet169Impl(
 DenseNet201Impl::DenseNet201Impl(
     int64_t num_classes,
     int64_t growth_rate,
-    std::vector<int64_t> block_config,
+    const std::vector<int64_t>& block_config,
     int64_t num_init_features,
     int64_t bn_size,
     double drop_rate)
@@ -202,7 +202,7 @@ DenseNet201Impl::DenseNet201Impl(
 DenseNet161Impl::DenseNet161Impl(
     int64_t num_classes,
     int64_t growth_rate,
-    std::vector<int64_t> block_config,
+    const std::vector<int64_t>& block_config,
     int64_t num_init_features,
     int64_t bn_size,
     double drop_rate)
diff --git a/torchvision/csrc/models/densenet.h b/torchvision/csrc/models/densenet.h
index 3ed6eba0837..7ec6e867698 100644
--- a/torchvision/csrc/models/densenet.h
+++ b/torchvision/csrc/models/densenet.h
@@ -26,7 +26,7 @@ struct VISION_API DenseNetImpl : torch::nn::Module {
   DenseNetImpl(
       int64_t num_classes = 1000,
       int64_t growth_rate = 32,
-      std::vector<int64_t> block_config = {6, 12, 24, 16},
+      const std::vector<int64_t>& block_config = {6, 12, 24, 16},
       int64_t num_init_features = 64,
       int64_t bn_size = 4,
       double drop_rate = 0);
@@ -38,7 +38,7 @@ struct VISION_API DenseNet121Impl : DenseNetImpl {
   DenseNet121Impl(
       int64_t num_classes = 1000,
       int64_t growth_rate = 32,
-      std::vector<int64_t> block_config = {6, 12, 24, 16},
+      const std::vector<int64_t>& block_config = {6, 12, 24, 16},
       int64_t num_init_features = 64,
       int64_t bn_size = 4,
       double drop_rate = 0);
@@ -48,7 +48,7 @@ struct VISION_API DenseNet169Impl : DenseNetImpl {
   DenseNet169Impl(
       int64_t num_classes = 1000,
       int64_t growth_rate = 32,
-      std::vector<int64_t> block_config = {6, 12, 32, 32},
+      const std::vector<int64_t>& block_config = {6, 12, 32, 32},
       int64_t num_init_features = 64,
       int64_t bn_size = 4,
       double drop_rate = 0);
@@ -58,7 +58,7 @@ struct VISION_API DenseNet201Impl : DenseNetImpl {
   DenseNet201Impl(
       int64_t num_classes = 1000,
       int64_t growth_rate = 32,
-      std::vector<int64_t> block_config = {6, 12, 48, 32},
+      const std::vector<int64_t>& block_config = {6, 12, 48, 32},
       int64_t num_init_features = 64,
       int64_t bn_size = 4,
       double drop_rate = 0);
@@ -68,7 +68,7 @@ struct VISION_API DenseNet161Impl : DenseNetImpl {
   DenseNet161Impl(
       int64_t num_classes = 1000,
       int64_t growth_rate = 48,
-      std::vector<int64_t> block_config = {6, 12, 36, 24},
+      const std::vector<int64_t>& block_config = {6, 12, 36, 24},
       int64_t num_init_features = 96,
       int64_t bn_size = 4,
       double drop_rate = 0);
diff --git a/torchvision/csrc/models/inception.cpp b/torchvision/csrc/models/inception.cpp
index 625de5c4ab2..002bf7ee2d1 100644
--- a/torchvision/csrc/models/inception.cpp
+++ b/torchvision/csrc/models/inception.cpp
@@ -49,7 +49,7 @@ InceptionAImpl::InceptionAImpl(int64_t in_channels, int64_t pool_features)
   register_module("branch_pool", branch_pool);
 }
 
-torch::Tensor InceptionAImpl::forward(torch::Tensor x) {
+torch::Tensor InceptionAImpl::forward(const torch::Tensor& x) {
   auto branch1x1 = this->branch1x1->forward(x);
 
   auto branch5x5 = this->branch5x5_1->forward(x);
@@ -76,7 +76,7 @@ InceptionBImpl::InceptionBImpl(int64_t in_channels)
   register_module("branch3x3dbl_3", branch3x3dbl_3);
 }
 
-torch::Tensor InceptionBImpl::forward(torch::Tensor x) {
+torch::Tensor InceptionBImpl::forward(const torch::Tensor& x) {
   auto branch3x3 = this->branch3x3->forward(x);
 
   auto branch3x3dbl = this->branch3x3dbl_1->forward(x);
@@ -115,7 +115,7 @@ InceptionCImpl::InceptionCImpl(int64_t in_channels, int64_t channels_7x7) {
   register_module("branch_pool", branch_pool);
 }
 
-torch::Tensor InceptionCImpl::forward(torch::Tensor x) {
+torch::Tensor InceptionCImpl::forward(const torch::Tensor& x) {
   auto branch1x1 = this->branch1x1->forward(x);
 
   auto branch7x7 = this->branch7x7_1->forward(x);
@@ -151,7 +151,7 @@ InceptionDImpl::InceptionDImpl(int64_t in_channels)
   register_module("branch7x7x3_4", branch7x7x3_4);
 }
 
-torch::Tensor InceptionDImpl::forward(torch::Tensor x) {
+torch::Tensor InceptionDImpl::forward(const torch::Tensor& x) {
   auto branch3x3 = this->branch3x3_1->forward(x);
   branch3x3 = this->branch3x3_2->forward(branch3x3);
 
@@ -185,7 +185,7 @@ InceptionEImpl::InceptionEImpl(int64_t in_channels)
   register_module("branch_pool", branch_pool);
 }
 
-torch::Tensor InceptionEImpl::forward(torch::Tensor x) {
+torch::Tensor InceptionEImpl::forward(const torch::Tensor& x) {
   auto branch1x1 = this->branch1x1->forward(x);
 
   auto branch3x3 = this->branch3x3_1->forward(x);
diff --git a/torchvision/csrc/models/inception.h b/torchvision/csrc/models/inception.h
index 46c224752cd..8d1d3707c82 100644
--- a/torchvision/csrc/models/inception.h
+++ b/torchvision/csrc/models/inception.h
@@ -24,7 +24,7 @@ struct VISION_API InceptionAImpl : torch::nn::Module {
 
   InceptionAImpl(int64_t in_channels, int64_t pool_features);
 
-  torch::Tensor forward(torch::Tensor x);
+  torch::Tensor forward(const torch::Tensor& x);
 };
 
 struct VISION_API InceptionBImpl : torch::nn::Module {
@@ -32,7 +32,7 @@ struct VISION_API InceptionBImpl : torch::nn::Module {
 
   InceptionBImpl(int64_t in_channels);
 
-  torch::Tensor forward(torch::Tensor x);
+  torch::Tensor forward(const torch::Tensor& x);
 };
 
 struct VISION_API InceptionCImpl : torch::nn::Module {
@@ -43,7 +43,7 @@ struct VISION_API InceptionCImpl : torch::nn::Module {
 
   InceptionCImpl(int64_t in_channels, int64_t channels_7x7);
 
-  torch::Tensor forward(torch::Tensor x);
+  torch::Tensor forward(const torch::Tensor& x);
 };
 
 struct VISION_API InceptionDImpl : torch::nn::Module {
@@ -52,7 +52,7 @@ struct VISION_API InceptionDImpl : torch::nn::Module {
 
   InceptionDImpl(int64_t in_channels);
 
-  torch::Tensor forward(torch::Tensor x);
+  torch::Tensor forward(const torch::Tensor& x);
 };
 
 struct VISION_API InceptionEImpl : torch::nn::Module {
@@ -62,7 +62,7 @@ struct VISION_API InceptionEImpl : torch::nn::Module {
 
   InceptionEImpl(int64_t in_channels);
 
-  torch::Tensor forward(torch::Tensor x);
+  torch::Tensor forward(const torch::Tensor& x);
 };
 
 struct VISION_API InceptionAuxImpl : torch::nn::Module {
diff --git a/torchvision/csrc/models/resnet.cpp b/torchvision/csrc/models/resnet.cpp
index ce9e95ecf6f..fe8059a15ec 100644
--- a/torchvision/csrc/models/resnet.cpp
+++ b/torchvision/csrc/models/resnet.cpp
@@ -28,7 +28,7 @@ BasicBlock::BasicBlock(
     int64_t inplanes,
     int64_t planes,
     int64_t stride,
-    torch::nn::Sequential downsample,
+    const torch::nn::Sequential& downsample,
     int64_t groups,
     int64_t base_width)
     : stride(stride), downsample(downsample) {
@@ -57,7 +57,7 @@ Bottleneck::Bottleneck(
     int64_t inplanes,
     int64_t planes,
     int64_t stride,
-    torch::nn::Sequential downsample,
+    const torch::nn::Sequential& downsample,
     int64_t groups,
     int64_t base_width)
     : stride(stride), downsample(downsample) {
diff --git a/torchvision/csrc/models/resnet.h b/torchvision/csrc/models/resnet.h
index 16a5f183f25..312ad7102da 100644
--- a/torchvision/csrc/models/resnet.h
+++ b/torchvision/csrc/models/resnet.h
@@ -36,7 +36,7 @@ struct VISION_API BasicBlock : torch::nn::Module {
       int64_t inplanes,
       int64_t planes,
       int64_t stride = 1,
-      torch::nn::Sequential downsample = nullptr,
+      const torch::nn::Sequential& downsample = nullptr,
       int64_t groups = 1,
       int64_t base_width = 64);
 
@@ -59,7 +59,7 @@ struct VISION_API Bottleneck : torch::nn::Module {
       int64_t inplanes,
       int64_t planes,
       int64_t stride = 1,
-      torch::nn::Sequential downsample = nullptr,
+      const torch::nn::Sequential& downsample = nullptr,
       int64_t groups = 1,
       int64_t base_width = 64);
 
diff --git a/torchvision/csrc/models/squeezenet.cpp b/torchvision/csrc/models/squeezenet.cpp
index cfb26ea58eb..96a9a1800d0 100644
--- a/torchvision/csrc/models/squeezenet.cpp
+++ b/torchvision/csrc/models/squeezenet.cpp
@@ -1,6 +1,5 @@
 #include "squeezenet.h"
 
-#include <limits>
 #include "modelsimpl.h"
 
 namespace vision {
diff --git a/torchvision/csrc/models/vgg.cpp b/torchvision/csrc/models/vgg.cpp
index f746acf6f34..73d32d98214 100644
--- a/torchvision/csrc/models/vgg.cpp
+++ b/torchvision/csrc/models/vgg.cpp
@@ -50,7 +50,7 @@ void VGGImpl::_initialize_weights() {
 }
 
 VGGImpl::VGGImpl(
-    torch::nn::Sequential features,
+    const torch::nn::Sequential& features,
     int64_t num_classes,
     bool initialize_weights) {
   classifier = torch::nn::Sequential(
diff --git a/torchvision/csrc/models/vgg.h b/torchvision/csrc/models/vgg.h
index cc9f98aea77..d3cd8917576 100644
--- a/torchvision/csrc/models/vgg.h
+++ b/torchvision/csrc/models/vgg.h
@@ -12,7 +12,7 @@ struct VISION_API VGGImpl : torch::nn::Module {
   void _initialize_weights();
 
   VGGImpl(
-      torch::nn::Sequential features,
+      const torch::nn::Sequential& features,
       int64_t num_classes = 1000,
       bool initialize_weights = true);