LayerLab

A visual neural network designer that allows you to design neural networks via drag and drop, with real-time parameter counting and PyTorch code generation.

Features

• 🎨 Drag-and-Drop Interface: Design networks by dragging layers onto a 2D canvas
• 🔗 Visual Connections: Draw connections between layers with visual feedback
• 📊 Real-Time Statistics: Live parameter and FLOP counting with highlighting for large layers
• ⚙️ In-Place Editing: Adjust hyperparameters directly on each node
• 🔄 Power-of-Two Snapping: Optional snapping to powers of two when adjusting parameters
• ✅ Validation: Automatic checking of layer compatibility with error/warning messages
• 🐍 PyTorch Export: Generate complete PyTorch model code
• 🔍 Workspace Zoom: Zoom in/out to manage complex architectures

Supported Layers

Input Layers

• Input1D (for sequential/tabular data)
• Input2D (for images)

Basic Layers

• Linear (fully connected)
• Conv2D (convolutional)
• MaxPool2D, AvgPool2D
• Dropout, Dropout2D
• BatchNorm1D, BatchNorm2D
• Flatten

Activation Functions

• ReLU, LeakyReLU, PReLU
• Sigmoid, Tanh
• Softmax, Softplus, Mish

Attention

• MultiheadAttention

Usage

Visit https://mikesha2.github.io/LayerLab/ to use the tool.

Or run locally:

1. Clone the repository
2. Open index.html in a web browser

Development

To minify the JavaScript:

npm install
npm run minify

License

MIT