line breaks inside a sentence
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. 4 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 when a line of text ends in the middle of a sentence, where the writing has been wrapped by hand or by a website. Turning it up did not produce a clear theme in the text it wrote.
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.
-
. - I was there too, but didn't⏎see you. I also will be working there from 3 to 5 tomorrow⏎7. When she⏎returns from Moscow, she'll tell us
-
any of the other auberges were situated on a piazza, is not⏎certain but shall be discussed later on.⏎Unfortunately, only three of⏎Cassar's seven auberges have survived. The German aub
-
that changes how the⏎cd command works. Normally when you try and cd⏎into a directory, it
-
greeting us at the top of the⏎steps. Adorning each of the vessels are the origami butterflies, both⏎male and female.⏎Several years ago I⏎wrote a Shinto scholar who teaches at a major American university. I asked
-
that it⏎bridges them. Using a straw, blow air underneath⏎the paper. This
-
," Professor Hacker said. "We want to understand⏎the physics behind it, so we can forecast it."⏎Russell White, of Byron Bay, was one⏎of two glider pilots who pioneered the art of soaring on a Morning Glory⏎
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 been thinking about myself. I have been thinking of me. I have been thinking of me. I have been thinking of me. I have been thinking of me. I have been thinking of me. I have been thinking of me
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
1922, have been preserved and donated to Olympic National Park by Doreen Taylor, the wife of Fannie's grandson. In her diaries Fannie describes the day to day activities of community life, such as interactions with road
-
strict discipline. The treatments were tested at UA’s Disruptive Behavior Clinic in advance of Salekin’s partnership with DYS.⏎The teens and pre-teens Salekin and his staff treat have a variety of behavioral problems, like oppositional defiance
-
story of a young prince who’s father recently died. Hamlets uncle Claudius marries his mother the queen and takes the throne. As the play is told Hamlet finds out his father was murdered by the recently crowned king. The theme that
-
you to know about how long it will take before your skin begins to redden while using a product that features it, such as sunscreen. (2)⏎How UPF Is Measured⏎Typ
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
- L3:131 suppresses
- L3:0 excites
- L3:155 suppresses
- L3:182 weakly suppresses
- L3:555 weakly suppresses
- L2:139 weakly excites
- L3:254 weakly suppresses
- L2:25 weakly suppresses
- L3:80 weakly suppresses
- L2:557 weakly excites
- L2:216 weakly excites
- L3:1034 weakly suppresses
- L2:270 weakly excites
- L2:464 weakly suppresses
- L3:1449 weakly suppresses
- L2:868 weakly suppresses line breaks inside lists
- L3:303 weakly excites
- L3:439 weakly suppresses
- L2:87 weakly suppresses
- L2:932 weakly excites
- L3:115 weakly suppresses
- L1:211 weakly excites
- L3:472 weakly suppresses
- L0:45 weakly excites
Feeds later layers
- L5:61 weakly excites
- L18:1012 weakly suppresses
- L7:264 weakly suppresses unclear: 'as' after a verb phrase
- L19:73 weakly excites
- L22:461 weakly suppresses
- L27:1866 weakly suppresses
- L11:2559 weakly suppresses
- L11:1130 weakly suppresses
- L18:42 weakly excites
- L23:268 weakly suppresses
- L17:370 weakly suppresses
- L20:2566 weakly suppresses
- L6:146 weakly excites
- L8:475 weakly suppresses
- L11:1571 weakly suppresses
- L15:11 weakly excites
- L9:1802 weakly suppresses
- L19:171 weakly excites
- L18:2141 weakly excites
- L22:215 weakly excites
- L22:1223 weakly suppresses
- L14:352 weakly excites
- L25:2529 weakly suppresses
- L18:70 weakly 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
- 而去
- 켓
- \Unit
- edian
- ęk
- nee
- ออกไป
- าศ
- LETTE
- ulle
Suppresses
- oling
- oled
- duino
- ruk
- efe
- '));⏎
- omo
- :');⏎
- ;</
- Dota