line break inside hard-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. 9 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 a line break that falls in the middle of a sentence, which happens when text was copied from a page with fixed-width lines. The strongest negative cases are hyphenated pieces inside web addresses. Turning it up did not give a clear topic.
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.
-
, and private foundations⏎had supported the development of a plethora of⏎curricula and approaches to instruction; these⏎led to such R&D organizations as the Biological⏎Sciences Curriculum Study, the Chemical Bond⏎Approach, and the Physical Science
-
Oder⏎River and assault Berlin. As this was happening, British and Canadian forces were slogging their way towards the Rhine (taking the majority of Western⏎Allied losses in
-
that under the British⏎Mercantilism and American imperialism in chapter 4. While the crops⏎could produce incredible yields when properly managed and with plenty of⏎water, doing so reduced the water table, added salt deposits to the⏎land
-
priestly hierarchy; no saints, martyrs, or even a⏎founder; no scripture or exegesis; no interest in education and art - in⏎fact, no positive, constructive aspect whatsoever. So vague and amorphous⏎
-
an event⏎known as a Lunar Eclipse, in which the Earth's circular⏎shadow can be seen to move across the face of the Moon. Because of the⏎orbital arrangements involved,
-
has said⏎that he would not have one Christian missionary in India if he rose to⏎power. And if the Zionist Jews became the governors of Palestine all missionary⏎work there
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.
-
Jeremy Rosen on Rav Lichtenstein is available from http://jeremyrosen.blogspot.co.uk/2015/08/rav-lichtenstein-zl.html⏎Your comment will be posted after it is approved.⏎Leave a Reply
-
://volcansvanuatueruptionsgb.blogspot.fr/2014/03/08032014-nevado-de-colima-cleveland.html and used this image from January 2nd. We
-
and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-cabinets-ensuring.html<|endoftext|>World
-
the stream they have investigated. If you are interested in becoming a⏎TEAM volunteer and helping educate Marylanders about our natural⏎resources, visit our website at⏎www.dnr.maryland.gov/education/teamdnr and fill out
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
- L7:74 strongly suppresses unclear: scattered mid-sentence tokens
- L8:380 strongly suppresses
- L4:156 strongly suppresses
- L4:32 strongly excites
- L7:1240 strongly excites
- L6:1374 suppresses
- L3:468 excites
- L8:538 excites
- L6:7 excites
- L5:1490 excites
- L8:7 suppresses
- L7:682 suppresses
- L5:1778 excites
- L8:3 suppresses
- L7:133 excites line break inside a sentence
- L4:138 suppresses
- L4:119 suppresses
- L3:225 suppresses
- L4:36 suppresses
- L8:4 suppresses
- L8:16 excites
- L2:126 suppresses
- L8:13 excites
- L7:122 suppresses
Feeds later layers
- L16:795 strongly suppresses
- L10:230 strongly excites
- L22:377 strongly suppresses
- L20:268 strongly suppresses
- L21:1164 strongly suppresses
- L10:102 strongly suppresses
- L14:568 strongly suppresses
- L16:317 strongly suppresses
- L14:2751 strongly suppresses
- L17:910 strongly suppresses
- L20:652 suppresses
- L17:1242 suppresses
- L19:1079 suppresses
- L20:320 suppresses
- L15:643 excites
- L12:1012 suppresses
- L15:985 suppresses line breaks inside wrapped text
- L10:2315 suppresses line break inside wrapped text
- L18:795 suppresses
- L15:1448 suppresses
- L25:135 excites
- L19:558 suppresses
- L19:2773 suppresses
- L15:337 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
- loquent
- moll
- {\
- entarios
- bilit
- {_
- izin
- `.
- adaptiveStyles
- 里面的
Suppresses
- ⏎
- ึ
- nels
- ⏎
- pund
- __(⏎
- éis
- ⏎
- AFE
- ',...⏎