Bonsai / Neuron Cloud
Ternary-Bonsai-2-2B · 1.58 bit · WebGPU
Neurons–
Token–
Speed–
StatusNot loaded
Fig. 01 Neuron cloud · 24 layersdrag to orbit · pinch to zoom · click a dot
Wake Bonsai to map its neurons.

Watch a 1.58-bit brain think.

Azimuth –
Elevation –
Shown –
Threshold +1.9σ
Far above usual
Above
Below
This word's path
Fig. 02 Word map · PCALayer 16 / 23

Each word's internal state at this layer, squashed to 2D. Words the model treats alike land close together. Low layers group spelling, high layers group meaning.

Fig. 03 Selected neuronσ above usual level
Click a dot in the cloud to follow one neuron.
Writeclick a word to replay its wave · click it again to unselect
Wake Bonsai, then press Generate.
Prompt words are grey, the model's own words are ink.
Next tokenwhat it considered at the selected word
Fig. 04 Surgeryclick blocks to switch them off · then press Apply
Weights off
0%
Loss now, nats
–
Intact, nats
–
Mixer blocks mix words together (A = full attention); MLP blocks hold the neurons above.

Switching a block off zeroes the scale shared by each group of 128 of its ternary weights, right in GPU memory, so the whole block counts as 0. Restore puts back the exact original bytes. Loss is the average surprise per token on a fixed paragraph of easy facts, so lower is smarter. The cloud keeps comparing against the intact model, so damage shows up as dots far from their usual level, and a switched-off MLP layer goes quiet (its ring turns orange).

Notes What you're looking at

The dots

Each layer of Bonsai has an MLP with 6,144 neurons. Showing all 147,456 would be a fog, so each layer shows the 192 that vary most on a short warm-up text. Their activations are copied off the GPU after every layer, for every word, and match PrismML's own llama.cpp to a correlation of 0.9999.

Where they sit

Height is the layer. Around each ring, a neuron's angle comes from which warm-up words make it fire (the first two principal components of its firing pattern), so neurons with similar tastes line up, even across layers. Distance from the centre is spread evenly so every dot is visible.

Colours, lines and the path

Colour is how far a neuron is from its own usual level on this word: orange is about 2 standard deviations or more above. Lines join each neuron to the neuron in the next layer it most often fires with, and turn orange when both fire. They show correlation, not wiring. The orange path follows the centre of the strongest firing in each layer.

Engine: PrismML's WebGPU runtime from their Talk app (Apache-2.0).Source · Test model (no GPU, fake activations)