Repository navigation
User Guide Scene Navigation
Scene Navigation refers to how TokenPrint transitions between different visual modes and how the camera moves between points of interest within those modes.
A clunky transition between exploring static tensors and watching live inference can break the user's mental model. Smooth scene navigation ensures you maintain spatial awareness of the Transformer Stack.
TokenPrint uses React to mount and unmount different Scene components inside the main AppShell canvas based on the selected mode in the Top Bar.
-
TensorCloudfor Architecture mode. -
GenerationScenefor Live Inference. -
WalkthroughScenefor the guided tour.
Camera navigation is handled by easing functions that interpolate the camera's position and look-at target over time.
- Architecture: Free-roam. The camera defaults to an overview of the point cloud.
-
Generation: Automated follow-mode. The camera glides down the
TransformerStackas thePlaybackEngineadvances through the layers. - Walkthrough: Chapter-gated. As you click "Next Chapter", the camera glides to specific curated locations (e.g., zooming in on the SwiGLU funnel or the KV Cache).
If you switch tabs or stress the GPU, WebGL may drop the rendering context. TokenPrint's SceneLoader.tsx acts as a safety boundary:
- It probes for WebGL availability.
- It automatically remounts a fresh canvas upon context loss.
- It displays a readable fallback message rather than crashing the browser.
flowchart TD
TopBar[Mode Selection] --> Store[Zustand Mode State]
Store --> Router{Mount Scene}
Router -- Architecture --> TensorCloud[Mount TensorCloud]
Router -- Generation --> GenerationScene[Mount GenerationScene]
Router -- Walkthrough --> WalkthroughScene[Mount WalkthroughScene]
GenerationScene --> Stack[Mount TransformerStack]
WalkthroughScene --> Stack
| Previous | Home | Next |
|---|---|---|
| Timeline | Home | Transformer Concepts |
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