§
Lab · 07 / Neuron Stream
v1 · live
Neuron Stream Lab.
A tiny transformer decoder that exposes its activations in real time — built to make the inner mechanics of a language model legible at a glance.
Output
idleNeuron Graph
0.00
Each circle is a neuron from a transformer layer. Brighter circles fire harder for the current token; lines trace which inputs and which neurons in adjacent layers feed each other most strongly.
Attention
0 tokensEach row is a layer; each column is a token in the running context. Brighter, faster pulses mark tokens the model is leaning on most when deciding what comes next.