number right after a product or name
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. 17 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 at the spot just before a version or part number, such as the 17 in 'Java SE 17' or the 10 in 'Windows 10'. Its negative side is on Chinese characters in mixed text. Turning it up gave unrelated text, so steering shows nothing clear.
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.
-
graph. You can see the full parent graph of cotangent, cot x in the figure.<|endoftext|>The steps are shown in IIS 10 on Windows 10 Enterprise, but they are also shown in IIS
-
Arabian lava field from the International Space Station:⏎Harrat Khaybar, Saudi Arabia is featured in this image photographed by an Expedition 16 crewmember on the International Space Station. The western half of the Arabian
-
, if it reaches orbit. If it doesn’t, it won’t be the first Landsat satellite to fail.⏎In 1993, Landsat 6 didn’t make it to orbit, leaving the nine-year-old Landsat
-
the board exams. Students can easily NCERT Solutions for Class 10 Maths Exercise 6.5 Chapter 6 Triangles from below.⏎CBSE Class 10 Maths Chapter 6 Triangles Exercise 6.5 Solutions⏎Topics
-
, fully updated and expanded⏎to cover Java SE 17 (JDK 17).⏎Whether you're an experienced pro or
-
know more, do more and remember more.⏎Progression of Skills⏎Our Music curriculum is planned so that there is a clear progression of skills from Nursery to Year 2 but within each year group we also meet the individual needs of all children by
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: it's a time that's going to be very important to be able to see it when it's on the next day. maybe 20 20 2 2 2 2 2
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
對 CNY 感興趣?⏎Early Currency in China⏎With a history of over 3
-
wind 雨 : rain 声 : sound⏎Meaning for Line 4 花 : flower
-
may, glass holders are oftentimes supplanted with plastic ones, in light of the fact that dissimilar to glass, plastic is impervious to affect, 食品包裝牌照 is lightweight and more functional. In any case,
-
-Western part of the Korean peninsula 3. The state of Paekche, because of the continuous pression exerted upon it by Koguryŏ 高句麗 4
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
- L4:53 excites blank slot before a number
- L9:7 weakly excites
- L13:53 weakly excites period ending a claim before elaboration
- L10:2401 weakly excites
- L14:768 weakly excites
- L5:5 weakly suppresses digit after a label like paper or type
- L16:215 weakly suppresses
- L15:114 weakly excites
- L10:2895 weakly excites
- L14:348 weakly suppresses
- L14:51 weakly excites colon that introduces a statement
- L11:31 weakly suppresses
- L16:572 weakly excites
- L15:82 weakly suppresses
- L1:2286 weakly excites
- L0:1882 weakly excites unclear: 'high tech' and mid-word pieces
- L6:249 weakly excites
- L16:1619 weakly excites
- L0:170 weakly suppresses
- L10:33 weakly suppresses
- L0:2086 weakly suppresses
- L16:353 weakly excites
- L13:923 weakly suppresses word right before a number or year
- L16:314 weakly excites
Feeds later layers
- L20:380 weakly excites
- L23:498 weakly suppresses
- L19:660 weakly suppresses
- L18:708 weakly excites
- L19:696 weakly excites
- L20:486 weakly excites
- L21:1159 weakly suppresses
- L18:558 weakly excites
- L21:642 weakly excites
- L20:1190 weakly suppresses
- L20:1504 weakly excites
- L19:1341 weakly excites "this" after a time phrase
- L18:632 weakly excites
- L19:183 weakly excites
- L20:374 weakly excites
- L20:225 weakly suppresses
- L22:895 weakly suppresses
- L18:871 weakly excites
- L24:2720 weakly suppresses
- L19:2582 weakly excites
- L19:1128 weakly suppresses
- L22:833 weakly suppresses
- L18:344 weakly excites
- L24:610 weakly suppresses connecting word in how-to steps
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
- away
- ordinary
- جاز
- olated
- Away
- Real
- vier
- omorphic
- ="{
- PartialView
Suppresses
- ))-
- ---</
- *>*
- preh
- neckline
- 大学毕业
- тин
- 做完
- assing
- ;</