lqdflying/transformer-explainer repository overview

⁠Transformer Explainer

Interactive visualization of how Transformer-based models like GPT work. Runs a live GPT-2 model in your browser.

⁠Quick Start

docker run -d --name transformer-explainer -p 8080:80 lqdflying/transformer-explainer:latest

Open http://localhost:8080⁠ in your browser.

⁠What's inside

• Node 20 + SvelteKit static build
• GPT-2 ONNX model (~600MB) — runs inference in-browser via ONNX Runtime Web
• GPT-2 tokenizer — bundled locally to avoid CORS issues with HuggingFace
• nginx — serves the static site with gzip compression and SPA fallback

⁠Features

• Type your own text prompt and click Generate to see next-token prediction
• Swap between 5 example prompts
• Adjust temperature and sampling strategy (top-k / top-p)
• Hover over vectors and attention maps to inspect values
• Step through transformer blocks with animations

⁠Ports

Change host port if 8080 is busy:

docker run -d --name transformer-explainer -p 3000:80 lqdflying/transformer-explainer:latest

⁠Environment Variables

⁠Build from source

git clone https://github.com/lqdflying/transformer-explainer.git
cd transformer-explainer
docker build -t transformer-explainer .
docker run -d --name te -p 8080:80 transformer-explainer

⁠Stop / Remove

docker stop transformer-explainer
docker rm transformer-explainer

⁠Image Size

~1.3 GB (includes ONNX model weights)

⁠Source

https://github.com/lqdflying/transformer-explainer⁠

⁠Original Project

https://poloclub.github.io/transformer-explainer⁠