Understanding the Interface

Getting Started

TokenPrint uses a docked-shell layout: a persistent top navigation bar, a central 3D canvas, and context-dependent side panels. The canvas is always present; panels and overlays appear based on the active mode.


Top navigation bar

The top bar contains, left to right:

  • TokenPrint logo — click to reset to the default view.
  • Mode buttons — Architecture, Generation, Walkthrough, Debugger.
  • Model selector — shows the currently loaded model name. Click to open the Model Explorer modal.
  • Report — generates a Model Health Report for the currently loaded model.
  • Docs — links to this documentation.

3D canvas

The central canvas renders a React Three Fiber scene. Camera controls:

InputAction
Left dragOrbit
Right drag / two-finger dragPan
Scroll / pinchZoom
Click an objectSelect and open Tensor Inspector
Double-clickFocus camera on clicked object

Camera bar

A unified camera control bar sits in the bottom transport bar and works identically in all modes: OVERVIEW, LAYER, OP, FOLLOW, a 3D/2D toggle, and TRACE. See Camera System for the full breakdown.


Right panel

The right panel is context-sensitive:

  • Architecture mode — Tensor Inspector showing selected tensor name, shape, dtype, and parameter count.
  • Generation mode — current op name, layer, KV phase, weight preview, top-k skyline, and generation controls.
  • Walkthrough mode — chapter explanation with real numbers, chapter navigation, model scale selector.
  • Debugger mode — tiled dashboard: breakpoints, flame graph, sentinels, watch expressions, layer table.

Keyboard shortcuts

KeyAction
SpacePlay / pause generation
← / →Step backward / forward one op
FReturn the camera to the overview position
TLive token-follow camera
1–4Switch mode (Architecture / Generation / Walkthrough / Debugger)
EscDeselect / close modal
RReset camera to default position