period then line break at end of line
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. 22 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 weakly on a full stop that is immediately followed by a line break, which usually happens at the end of a sentence or list item in web text. It is a weak signal compared with the strongest neurons. Steering did not produce a clear theme.
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.
-
80 p. :|⏎|Number of Pages||99|⏎|2||WCC exchange -- 2|⏎nodata File Size: 5MB.⏎Yet for many in the Roman Catholic church in Latin America, not to mention in
-
Arabic. Number of native speakers: 221 million. …⏎What is the best app to learn Hebrew?⏎Online Hebrew courses⏎- Learn Hebrew Easily.⏎- Mondly Languages.⏎- i<|endoftext|>As the citizens of the United States follow
-
AND BENEFICIATION PRACTICES . In most other cases, the ore is crushed and/or ground, prior to leaching or flotation. Each of the.⏎10 OPERATIONS: ORE PROCESSING - Midas Gold⏎Ore
-
decreased by nearly two-thirds.⏎“Tsunami Early Warning System Moves Ahead at UN Workshop,” UN News Service, 14 June 2006.⏎Link: http://www.un.org/apps/news/story.asp?NewsID=1
-
peak of our region’s fire emergency, somebody rang our Rural Fire Service (RFS) headquarters and complained: ‘Why haven’t you put the fires out?’.⏎There are parallels between the current pandemic, the recent fire emergency and ‘wildfires
-
the Murray B. Sachs Professor in the Department of Biomedical Engineering at The Johns Hopkins University Schools of Engineering and Medicine and the project’s principal investigator… Continue reading.⏎Natalia Trayanova, a professor
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: Is it possible for thesitiess tos
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
its vestiges.⏎Long before the liberal revolution of Europe dealt the most effective political blows against feudal power in the 17th⏎and 18th century, a cultural revolution took shape in the Renaissance which
-
air in buildings. Its use involves the construction of vertical ventilation channels, which are designed to carry away warm air from the upper layers of the interior to the outside⏎Its operation is based on the pressure difference generated in the ventilation duct under appropriate atmospheric
-
and rocket projectiles respectively.⏎Mahogany Wood. Wingspan 19 inches, Length 19 inches.<|endoftext|>A random variable X has the cumulative distribution function⏎Calculate the variance of X.⏎The answer specifies that the density function is⏎I
-
It was named after George C. Marshall, who was the US Secretary of State. Thousands of millions of dollars were provided for rebuilding cities, roads, industries etc⏎Definition from the Longman Dictionary of Contemporary English Advanced Learner's Dictionary.⏎Dictionary
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
- L17:829 suppresses
- L10:84 suppresses
- L8:32 excites
- L18:429 suppresses
- L19:187 suppresses
- L16:359 suppresses
- L13:10 excites
- L10:33 suppresses
- L15:123 suppresses
- L21:173 excites
- L1:1 excites
- L16:170 suppresses
- L18:5 suppresses
- L6:67 excites
- L9:25 excites
- L18:227 suppresses
- L21:130 excites
- L14:56 suppresses document ends and list line breaks
- L5:331 suppresses
- L18:48 weakly suppresses
- L18:124 weakly suppresses
- L6:4 weakly suppresses
- L9:7 weakly excites
- L14:57 weakly suppresses
Feeds later layers
- L25:577 strongly excites
- L23:393 suppresses
- L27:2408 suppresses
- L24:1512 excites
- L26:1008 suppresses
- L25:535 excites commas inside number lists
- L24:318 suppresses
- L24:2198 excites closing parentheses in lists
- L24:2281 weakly excites
- L26:362 weakly excites
- L27:740 weakly excites
- L26:505 weakly excites unclear: opening quote marks after a line break
- L26:402 weakly suppresses
- L25:396 weakly suppresses
- L23:1613 weakly suppresses
- L23:585 weakly suppresses
- L23:703 weakly suppresses
- L25:2123 weakly excites
- L27:2630 weakly excites
- L27:1007 weakly excites
- L27:2268 weakly excites
- L23:806 weakly suppresses
- L26:47 weakly suppresses
- L26:2753 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
- elda
- TING
- living
- ��
- ught
- living
- 脔
- world
- BED
- quito
Suppresses
- s
- sar
- sil
- ات
- 员
- sie
- enticate
- acters
- sand
- sse