NeuralFlow-Core is a high-performance, lightweight deep learning framework designed for efficiency and flexibility. It provides a seamless interface for building and training complex neural networks, with a focus on speed and scalability.
- High Performance: Optimized for both CPU and GPU execution, ensuring fast training and inference.
- Lightweight Architecture: Minimal dependencies and a clean codebase for easy integration and customization.
- Flexible API: Intuitive Python and C++ APIs for building diverse neural network architectures.
- Scalability: Designed to handle large-scale datasets and complex models with ease.
- Extensibility: Easily add new layers, activation functions, and optimization algorithms.
- Python 3.7+
- C++ Compiler (GCC 7+, Clang 5+, or MSVC 2017+)
- CMake 3.10+
- (Optional) CUDA 10.0+ for GPU acceleration
git clone https://github.com/FunctionFlow1/NeuralFlow-Core.git
cd NeuralFlow-Core
mkdir build && cd build
cmake ..
make -j$(nproc)
sudo make installimport neuralflow as nf
# Define a simple neural network
model = nf.Sequential([
nf.layers.Dense(128, activation='relu', input_shape=(784,)),
nf.layers.Dropout(0.2),
nf.layers.Dense(10, activation='softmax')
])
# Compile the model
model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
# Train the model
model.fit(x_train, y_train, epochs=10, batch_size=32)
# Evaluate the model
loss, accuracy = model.evaluate(x_test, y_test)
print(f"Test Accuracy: {accuracy:.4f}")We welcome contributions from the community! Please read our Contributing Guidelines for more information.
NeuralFlow-Core is released under the MIT License.