document boundary, code keywords next
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 27 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the marker that ends one document and starts another. When it is turned up, the model tends to write programming words like 'class', 'static' and 'public'.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
old<|endoftext|>Immune cells in the brain have surprising influence on sexual behavior⏎Relatively uncommon
-
data from TC1.<|endoftext|>The concept of welfare state⏎We all know that most of the modern democratic countries
-
locate.<|endoftext|>Science Fair Project Encyclopedia⏎The World Factbook⏎The World Factbook is an
-
needed around planes en route to keep them safely separated.<|endoftext|>FEDERAL Hill House is a squat building in central Providence, within earshot
-
/definition/acanthosis nigricans>.<|endoftext|>Dronacharya was the teacher of both the Kauravas and Pand
-
to like.<|endoftext|>The late Dominica-born writer Jean Rhys, best known for her novel “
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: perhaps you may be able to avoid criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism criticism
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
’ Google), you’re sure to spot subtle references that talk about ‘timeless content’.⏎No, this doesn’t necessarily mean that you have to whip up a Silmarillion every time you open MS Word – all you need to do is
-
a strand.⏎Very few organic molecules can with stand high temperatures for very⏎long. Heat provides the activation energy to break covalent bonds and⏎denature/destroy molecular structure. Proteins are particularly sensitive.⏎There are exceptions, as in
-
, that active or hyperactive kids moving in rhythm with the group, such as in walking meditation, derive a calming effect. 11. Meditation is non-competitive, and each can participate within the limits of their own ability, with adaptations
-
. Instances of the class Class represent classes and interfaces in a running Java application. If we do not explicitly write a constructor for a class, the Java compiler builds a default constructor for that class. so now we can use this to create objects.
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L19:542 suppresses
- L26:120 suppresses
- L22:633 suppresses
- L23:1647 excites
- L22:39 suppresses
- L14:67 suppresses
- L18:284 suppresses
- L15:524 excites
- L17:2265 excites
- L19:1372 suppresses
- L21:155 suppresses
- L24:721 suppresses
- L20:205 excites
- L26:399 suppresses
- L26:612 excites end-of-text boundary tokens
- L17:95 suppresses
- L14:1636 excites
- L23:81 excites
- L22:1816 weakly excites
- L10:38 weakly suppresses
- L9:18 weakly suppresses text boundary after a document break
- L21:162 weakly excites
- L21:131 weakly excites
- L17:35 weakly excites
Feeds later layers
None: this is the last layer.
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- class
- static
- static
- class
- public
- public
- struct
- function
- template
- auto
Suppresses
- ее
- ierrez
- вен
- ämpfe
- 閣
- -care
- 회의
- criticism
- ậm
- unately