Repository navigation
FAQ
Frequently asked questions about running and using TokenPrint.
The frontend cannot reach the backend.
- Ensure the backend is running:
python -m uvicorn app.main:app --port 8000 - Ensure it is on exactly port
8000. - Check your terminal for PyTorch loading errors.
The default model (Qwen/Qwen2.5-0.5B-Instruct) must be downloaded from HuggingFace (about 1GB) on the first run. Even when cached, loading the weights into RAM/VRAM takes 5-15 seconds depending on your hardware.
Yes, but it depends on your hardware.
TokenPrint loads models in float32 with eager attention to capture precise mathematical states. A 7B model requires massive amounts of RAM (often >30GB) to run in this mode. If you run out of memory, the backend will crash.
No, not currently.
You can drag and drop a .gguf to view its static structure in the Architecture Explorer, but Live Inference requires the PyTorch engine. Support for in-browser WebGPU inference for GGUF files is on the roadmap.
Your browser likely dropped the WebGL context due to GPU strain, or hardware acceleration is disabled in your browser settings. TokenPrint requires a modern browser with WebGL enabled.
TokenPrint aims to show exactly how math produces output. Stochastic sampling (Temperature, Top-P) introduces randomness that makes debugging specific behaviors difficult. A Sampling Playground for exploring temperature is planned.
| Previous | Home | Next |
|---|---|---|
| Roadmap | Home | Contributing |
TokenPrint Wiki | Home | Getting Started | User Guide | Concepts | Architecture | API | GitHub
TokenPrint is an interactive 3D visualization platform for exploring transformer architectures, tensors, and real-time LLM inference.
- User Guide
- Architecture Explorer
- Live Inference
- Tensor Inspector
- Camera Controls
- HUD
- Timeline
- Scene Navigation
- Transformer Concepts
- Tokenization
- Embeddings
- Positional Encoding
- RoPE
- Multi-Head Attention
- KV Cache
- LayerNorm
- Feed Forward Network
- Residual Connections
- Logits
- Softmax
- Sampling
- Autoregressive Generation
- Visualization System
- Scene Graph
- Geometry
- Materials
- Animation System
- Color Mapping
- Camera System
- Interaction Model
- Architecture
- Overall Architecture
- Frontend
- Backend
- WebSocket Protocol
- Data Pipeline
- Renderer
- Event System