diff --git a/README.md b/README.md
index ddb35b9..a3c7907 100644
--- a/README.md
+++ b/README.md
@@ -4,20 +4,7 @@
 
 Convolutional Neural Network inference library running on CUDA.
 
-## Features
-
-- [x] Input layer
-- [x] Dense (fully-connected) layer
-- [x] Conv2d layer
-- [x] Max pooling
-- [x] Average pooling
-- [x] Concat layer
-- [x] Sigmoid activation
-- [x] ReLU activation
-- [x] Softmax activation
-- [x] Load weights from file 
-
-## Usage
+## Quickstart Guide
 
 **requirements**
 - [cmake](https://cmake.org/)
@@ -94,7 +81,7 @@ float* MyModel::predict(const float* input) {
 CUDANet uses format similar to safetensors to load weights and biases.
 
 ```
-[int64 header size, header, tensor values]
+[u_short version, u_int64 header size, header, tensor values]
 ```
 
 where `header` is a csv format
@@ -103,6 +90,4 @@ where `header` is a csv format
 <tensor_name>,<tensor_size>,<tensor_offset>
 ```
 
-To load weights call `load_weights` function on Model object.
-
-To export weights from pytorch you can use the `export_model_weights` function from `tools/utils.py`  script
\ No newline at end of file
+To load weights call `load_weights` function on Model object. To export weights from pytorch you can use the `export_model_weights` function from `tools/utils.py`  script. Currently only float32 weights are supported
\ No newline at end of file
diff --git a/src/model/model.cpp b/src/model/model.cpp
index a8530f6..e992f14 100644
--- a/src/model/model.cpp
+++ b/src/model/model.cpp
@@ -66,7 +66,15 @@ void Model::loadWeights(const std::string& path) {
         return;
     }
 
-    int64_t headerSize;
+    u_short version;
+    file.read(reinterpret_cast<char*>(&version), sizeof(version));
+
+    if (version != 1) {
+        std::cerr << "Unsupported model version: " << version << std::endl;
+        return;
+    }
+
+    u_int64_t headerSize;
     file.read(reinterpret_cast<char*>(&headerSize), sizeof(headerSize));
 
     std::string header(headerSize, '\0');
@@ -110,7 +118,7 @@ void Model::loadWeights(const std::string& path) {
     for (const auto& tensorInfo : tensorInfos) {
         std::vector<float> values(tensorInfo.size);
 
-        file.seekg(sizeof(int64_t) + header.size() + tensorInfo.offset);
+        file.seekg(sizeof(version) + sizeof(headerSize) + header.size() + tensorInfo.offset);
         file.read(reinterpret_cast<char*>(values.data()), tensorInfo.size * sizeof(float));
 
         if (layerMap.find(tensorInfo.name) != layerMap.end()) {
diff --git a/test/resources/model.bin b/test/resources/model.bin
index d58f27a..2f2ff22 100644
Binary files a/test/resources/model.bin and b/test/resources/model.bin differ
diff --git a/tools/utils.py b/tools/utils.py
index 6c4f689..95da2b8 100644
--- a/tools/utils.py
+++ b/tools/utils.py
@@ -16,6 +16,7 @@ def print_cpp_vector(vector, name="expected"):
 def export_model_weights(model: torch.nn.Module, filename):
     with open(filename, 'wb') as f:
 
+        version = 1
         header = ""
         offset = 0
         tensor_data = b""
@@ -33,7 +34,8 @@ def export_model_weights(model: torch.nn.Module, filename):
             tensor_data += tensor_bytes
 
         f.seek(0)
-        f.write(struct.pack('q', len(header)))           
+        f.write(struct.pack('H', version))
+        f.write(struct.pack('Q', len(header)))           
         f.write(header.encode('utf-8'))
         f.write(tensor_data)