diff --git a/alexnet/alex.cpp b/alexnet/alex.cpp
index 17f2a10..f5e649b 100644
--- a/alexnet/alex.cpp
+++ b/alexnet/alex.cpp
@@ -233,7 +233,7 @@ int main(int argc, char** argv)
         APIToModel(1, &modelStream);
         assert(modelStream != nullptr);
 
-        std::ofstream p("alexnet.engine");
+        std::ofstream p("alexnet.engine", std::ios::binary);
         if (!p)
         {
             std::cerr << "could not open plan output file" << std::endl;
diff --git a/inception/inceptionv4/inception_v4.cpp b/inception/inceptionv4/inception_v4.cpp
index 80ea80c..0562fde 100644
--- a/inception/inceptionv4/inception_v4.cpp
+++ b/inception/inceptionv4/inception_v4.cpp
@@ -44,7 +44,7 @@ namespace trtx {
         builder -> destroy();
 
         // write serialized engine to file
-        std::ofstream trtFile(mParams.trtEngineFile);
+        std::ofstream trtFile(mParams.trtEngineFile, std::ios::binary);
         if(!trtFile){
             std::cerr << "Unable to open engine file." << std::endl;
             return false;
@@ -193,26 +193,26 @@ namespace trtx {
         // Engine requires exactly IEngine::getNbBindings() number of buffers.
         assert(mEngine -> getNbBindings() == 2);
         void* buffers[2];
-    
+
         // In order to bind the buffers, we need to know the names of the input and output tensors.
         // Note that indices are guaranteed to be less than IEngine::getNbBindings()
         const int inputIndex = mEngine->getBindingIndex(mParams.inputTensorName);
         const int outputIndex = mEngine->getBindingIndex(mParams.outputTensorName);
-    
+
         // Create GPU buffers on device
         CUDA_CHECK(cudaMalloc(&buffers[inputIndex], batchSize * 3 * mParams.inputH * mParams.inputW * sizeof(float)));
         CUDA_CHECK(cudaMalloc(&buffers[outputIndex], batchSize * 1000 * sizeof(float)));
-    
+
         // Create stream
         cudaStream_t stream;
         CUDA_CHECK(cudaStreamCreate(&stream));
-    
+
         // DMA input batch data to device, infer on the batch asynchronously, and DMA output back to host
         CUDA_CHECK(cudaMemcpyAsync(buffers[inputIndex], input, batchSize * 3 * mParams.inputH * mParams.inputW * sizeof(float), cudaMemcpyHostToDevice, stream));
         mContext->enqueue(batchSize, buffers, stream, nullptr);
         CUDA_CHECK(cudaMemcpyAsync(output, buffers[outputIndex], batchSize * 1000 * sizeof(float), cudaMemcpyDeviceToHost, stream));
         cudaStreamSynchronize(stream);
-    
+
         // Release stream and buffers
         cudaStreamDestroy(stream);
         CUDA_CHECK(cudaFree(buffers[inputIndex]));
@@ -232,4 +232,5 @@ namespace trtx {
 
         return true;
     }
-}
\ No newline at end of file
+}
+
diff --git a/lenet/lenet.cpp b/lenet/lenet.cpp
index 6df3c8e..5fba240 100644
--- a/lenet/lenet.cpp
+++ b/lenet/lenet.cpp
@@ -221,7 +221,7 @@ int main(int argc, char** argv)
         APIToModel(1, &modelStream);
         assert(modelStream != nullptr);
 
-        std::ofstream p("lenet5.engine");
+        std::ofstream p("lenet5.engine", std::ios::binary);
         if (!p)
         {
             std::cerr << "could not open plan output file" << std::endl;