DLAI (Deep Learning Algorithm Implementation)

DLAI is a streamlined repository focused on implementing core deep learning algorithms from scratch in C/C++. It is designed primarily for educational purposes and optimized for deployment on embedded systems and microcontrollers, supporting TinyML/EmbeddedAI applications. This repository serves as a foundational reference for understanding core deep learning principles and exploring TinyML applications.

---

🧠 Current State

• The project currently supports conversion of sequential neural networks from the PyTorch library only.
• It uses torch.save() to save the full model, and loads it using:

  torch.load(pth_file_path, weight_only=False)

✅ Layers Currently Supported

0. Linear (Fully Connected) Layer
1. ReLU Layer

Additional layers and features are actively under development.

---

🚀 Features

• Algorithm Implementations: Basic neural network layers implemented from scratch.
• Embedded Systems Focus: Optimized for TinyML and embedded system deployment.
• Educational Resource: A practical reference for learning deep learning fundamentals and embedded deployment.

---

🛠️ Getting Started

1. Clone the Repository

git clone https://github.com/Mat-thias/dlai.git

2. Navigate to the Project Directory

cd dlai

3. Convert a PyTorch Sequential Model to a C Header File

Run the script located in dlai/Models:

python3 dlai/Models/convert_sequential_model_to_c.py <model_file> <model_name> <input_shape> [output_dir]

Note: It is reocommended to create a virtual environment and install the packages in dlai/Models/venv_requirement.txt

Example

python3 dlai/Models/convert_sequential_model_to_c.py dlai/examples/sine_model/sine_model.pth sine_model "(1,1)" dlai/examples/sine_model/

---

⚙️ Example Usage

Layer *graph[LAYER_LEN];
float workspace[MAX_WORKSPACE_SIZE];

Sequential model(sine_model, sine_model_len, graph, LAYER_LEN, workspace, MAX_WORKSPACE_SIZE);

float *input = model.input;
float *output = model.output;

*input = (float)i * 2 * 3.141 / 360;
model.predict();
Serial.println(*output);

---

📂 Explore the Implementations

• Contains C/C++ source code for supported layers.
• Each implemented component includes inline documentation.

---

🤝 Contributing

We welcome contributions to expand supported layers and improve the tool!

1. Fork the Repository on GitHub.

2. Clone Your Fork:

   git clone https://github.com/your-username/dlai.git

3. Create a Feature Branch:

   git checkout -b feature/your-feature-name

4. Make Your Changes.

5. Commit Your Changes:

   git commit -m "Add your commit message"

6. Push to Your Fork:

   git push origin feature/your-feature-name

7. Open a Pull Request to the main repository.