line breaks inside wrapped text
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. 15 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 mostly fires on line breaks or the word piece right before one, especially inside poems and wrapped paragraphs. It seems to sense where a line of text is about to end. It tends to go negative right before a list or a colon. Turning it up gave off-topic rambling.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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.
-
who already have high-incomes; almostnone of it goes to the poorest Americans who need the most helpbuilding the financial assets that can lead to long-term economicopportunities and security.<|endoftext|>Modern methods of time management incorporate various ways of recording time
-
-eastern Namibia and South Africa, specifically in the Kruger⏎National Park, KwaZulu-Natal and the Eastern Cape. It prefers woodland dominated⏎by a broad-leaved trees such as bushwillows (Combretum),
-
, at sacred-texts.com⏎LONG is the mountain pheasant's tail⏎That curves down in its flight;⏎But longer still, it seems to me,⏎Left in my lonely plight,⏎Is this unending night.⏎The writer was a
-
I could not take,⏎Although so well inclined.⏎The rebels came with savage yells,⏎A Yankee they had found,⏎With curses bid me “face about,”⏎And march away to town.⏎They marched me down with many more,⏎Through mud
-
deficiency, its clinical manifestations and⏎the possible repercussions for Australia’s refugee population.⏎Full blood count and blood film, iron studies and haemoglobinopathy studies are⏎routinely performed for newly arrived refugees in Australia. At the Migrant Health
-
this spring morning, and missed the dawn, From everywhere I heard the cry of birds. That night the sound of wind and rain had come, Who knows how many petals then had fallen?<|endoftext|>At EdPlace we are passionate about
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: I have a but I have I have a This is the same as...? Wait, that's the same... Wait, I'm getting Wait, I'm Wait, I'm I'm Wait Wait
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
time if the child does not get any help. Some signs a child is having difficulties can be:⏎Behavior problems are usually easy to see. They can include ongoing:⏎- aggression or bullying⏎- refusal to do what they are asked
-
-blame/failure/low self-esteem Sadness/depression: loss . In Building Self-Esteem in Children Who Are Adopted or Fostered, Dr. Sue offers simple and practical advice to those supporting children aged 7+ to help
-
People can be exposed to RF radiation from both natural and man-made sources. Natural sources include:⏎Man-made RF radiation is used for many different things, such as:⏎Most people are exposed to much lower levels of man-made RF radiation every day due
-
Biff should be making good money and blames his son’s failure on his laziness. But it is Willy who has se... ... middle of paper ... ...d a the country”
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
- L10:230 suppresses
- L9:234 suppresses line break inside hard-wrapped text
- L10:2315 excites line break inside wrapped text
- L7:239 excites
- L10:494 suppresses
- L7:133 suppresses line break inside a sentence
- L7:1240 suppresses
- L13:2091 suppresses
- L7:2925 suppresses
- L13:309 excites
- L9:1238 excites
- L10:258 excites
- L12:601 weakly excites question mark or line break after an FAQ heading
- L12:2216 weakly excites
- L9:511 weakly suppresses
- L8:2432 weakly suppresses line breaks in lists and wrapped text
- L3:79 weakly excites
- L8:380 weakly suppresses
- L12:1012 weakly excites
- L14:2751 weakly suppresses
- L9:246 weakly suppresses
- L9:161 weakly excites
- L9:909 weakly suppresses
- L2:140 weakly excites "one" after an ordinal or superlative
Feeds later layers
- L17:1242 excites
- L16:442 excites
- L17:554 excites
- L16:822 excites
- L16:317 suppresses
- L20:1467 weakly suppresses
- L17:916 weakly excites
- L19:435 weakly suppresses
- L24:1197 weakly excites
- L25:239 weakly suppresses
- L24:406 weakly suppresses
- L17:1612 weakly excites
- L23:1763 weakly suppresses
- L17:1564 weakly excites
- L16:2249 weakly suppresses
- L17:2056 weakly suppresses
- L19:323 weakly suppresses
- L19:970 weakly excites
- L20:1571 weakly excites
- L22:667 weakly suppresses
- L17:1342 weakly suppresses
- L24:1619 weakly suppresses
- L26:144 weakly excites
- L17:2875 weakly suppresses
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
- /*.
- .gif
- acionales
- etration
- /releases
- igans
- �
- /loader
- .ogg
- .bmp
Suppresses
- auto
- matic
- WEST
- aut
- Auto
- InitialState
- AUTO
- 出来
- 提供优质
- �