A lightweight, beginner-friendly causal Transformer language model written from scratch in PyTorch, paired with a modern web dashboard designed in the style of ChatGPT/Claude. Runs entirely on local CPU, requiring less than 500 MB RAM.
- BPE Tokenizer Visualizer: Visualizes Byte-Pair Encoding splits and maps text characters to color-coded token IDs dynamically.
- Architecture Configuration: Customize parameters between Nano (1M params), Micro (10M params), or Mini (50M params) presets.
- Live CPU Training Monitor: Train the model on the tiny Shakespeare dataset in your browser with real-time Chart.js loss graphs, logs, and play/pause controls.
- GPT/Claude Chat Playground: Chat with the trained model or use the untrained fallback. Supports temperature and Top-K controls.
- Modern Switchable Themes: Toggle between White (Light), Grey (Slate), Yellow (Warm Beige), and Black (OLED) color layouts.
-
Setup Environment:
python -m venv venv # Activate on Windows: venv\Scripts\activate # Activate on Linux/Mac: source venv/bin/activate
-
Install Dependencies:
# Install PyTorch CPU first (lighter download size) pip install torch --index-url https://download.pytorch.org/whl/cpu # Install other dependencies pip install -r requirements.txt
-
Start the Web Server:
python web_server.py
Open your browser and navigate to http://127.0.0.1:8000!