end-of-document boundary token
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. 18 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 responds to the invisible marker the training text uses to show that one web page has ended and another has begun. It is only weakly active, and turning it up or down mostly made the model's output messy or full of asterisks.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).
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.
-
and improve communication will get them ready for job interviews. Long-term tracking, counselling, and support will be<|endoftext|>In this video, students learn that the Exxon Valdez oil spill in Alaska in
-
Engineering Career Development, check this out: http://engineeringcareercoach.com/community/⏎We would love to hear<|endoftext|>from The American Heritage® Dictionary of the English Language, 4th Edition⏎
-
utf8mb4) bytes per character can be used. So if one specifies VARCHAR(50) CHARSET utf8mb4, …[Read more]<|endoftext|>To actually reduce pounds in a healthier manner, it is perfect to burn fat gradually
-
measures for tobacco control in the Region based on the provisions of the Convention. The Meeting of the Advisory Committee (ACM), held at WHO/SEARO<|endoftext|>A finely modelled Ancient Roman cast bronze all-round statuette in the form
-
⏎July 08, 2009 |<|endoftext|>Computer Science Concentration for Students Matriculated After Jan. 1,
-
and resource for Scrabble & Words With Friends players.<|endoftext|>The Aioi Bridge is a T-shaped bridge that was originally targeted by American
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: **** ***** **** **** **** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
|⏎Koothus in Silapathikaram⏎by Subashini Pathmanathan⏎Ancient Tamils called their dances by
-
Coahuila y Nuevo León, Cuartel General en Monterey, Enero 3 de 1841, broadside, photostat.⏎3. Mariano Arista<|endoftext|>Dating photos relies on a combination of:⏎-
-
ментарии при наличии в них нецензурной брани, оскорблений или спама<|endoftext|>Conflict and Solidarity in Mudejar Society⏎The activity of the industrious
-
презентации Презентация london-history-1216214947983326-8 по слайдам⏎2008 by Mag. Andreas Hofer and the 4 DHW
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
- L17:1340 excites
- L17:1176 suppresses
- L15:65 excites
- L2:49 suppresses
- L5:78 weakly suppresses
- L2:2439 weakly suppresses
- L2:549 weakly suppresses
- L2:70 weakly excites
- L5:158 weakly suppresses
- L1:2233 weakly excites
- L17:954 weakly excites
- L17:581 weakly excites
- L5:169 weakly excites
- L5:110 weakly suppresses
- L16:270 weakly excites
- L11:131 weakly excites
- L1:2261 weakly suppresses
- L8:453 weakly suppresses
- L12:2728 weakly suppresses
- L17:1568 weakly suppresses
- L12:1818 weakly excites
- L16:1297 weakly excites unclear: document ends and answer colons
- L17:247 weakly suppresses
- L13:1174 weakly suppresses
Feeds later layers
- L26:156 strongly excites
- L24:96 strongly suppresses
- L24:303 strongly excites
- L23:194 strongly suppresses
- L26:288 strongly excites
- L26:16 strongly suppresses
- L25:207 strongly excites
- L21:1150 strongly excites
- L22:261 strongly excites
- L19:2077 suppresses
- L25:1274 excites
- L26:1180 excites
- L27:136 excites
- L27:179 suppresses
- L19:1105 excites
- L26:133 excites
- L27:445 excites
- L20:561 excites end of document marker
- L27:256 suppresses
- L25:144 suppresses
- L22:734 suppresses
- L27:1065 excites
- L26:640 suppresses
- L27:309 excites
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
- ('');⏎⏎
- ("");⏎⏎
- !⏎⏎
- ''⏎⏎
- ;⏎⏎
- '';⏎⏎
- [];⏎⏎
- ();⏎⏎
- !)⏎⏎
- <>();⏎⏎
Suppresses
- **⏎
- **⏎
- ****⏎
- **
- |⏎
- **
- *⏎
- *⏎
- |⏎
- .*⏎