blank spot right before a number
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. 14 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 the space just before a number or amount, such as the gap before '1.25mm', '70mm' or '1.4 billion', where the next word is likely to be a figure. Turning it up made the model's writing drift into repeated or odd number-heavy text, so the steering shows no 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.
-
and Persuasion with Nothing But the Truth⏎|Grades||6 – 8|⏎|Lesson Plan Type||Standard Lesson|⏎|Estimated Time||Five 50-minute sessions|⏎MATERIALS AND TECHNOLOGY⏎Copies of Av
-
.25 x 1 Meter. The M8 (8mm) is the rod's diameter, the 1.25 means that there is a 1.25mm distance between each thread, and the 1 meter is
-
the Gemini IV mission during orbit no. 19, on June 4, 1965, with a Hasselblad camera and a 70mm lens. The Gemini IV crew — astronauts Jim McDivitt and
-
akes recommend getting from 45 to 65 percent of daily calories - from 900 to 1,300 calories in a typical 2,000-calorie diet
-
<|endoftext|>The New International Encyclopædia/Kinkajou⏎|←Kinic acid||The New International Encyclopædia⏎|Edition of 1905. See also Kinkajou on Wikipedia, and the
-
PN Bridge Rectifier Diode (x4)⏎- 1N5819 Schottky Diode⏎- Resistor (2MΩ, 560Ω, 1KΩ, 10Ω,
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: my own experience, my own experience, my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
017 WHSmith and its suppliers.⏎WHSmith High Street Limited Greenbridge Road, Swindon, Wiltshire, United Kingdom, SN3 3
-
an appointment with one of our tutors. If you have any questions, please contact us at firstname.lastname@example.org or 410.837.5383.<|endoftext|>Microsoft co-founder Paul Allen has shown off the “
-
155. 50-58. 10.1016/j.biocon.2012.05.013Full text not available from this repository.⏎Although grasslands are crucial habitats
-
Drive, Headington, Oxford, OX3 8QQ⏎Back to previous page<|endoftext|>A waterspout is a cyclonic wind
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
- L11:226 suppresses
- L10:220 suppresses hyphen inside compound words
- L8:350 suppresses blank space before a year
- L4:53 excites blank slot before a number
- L12:1467 suppresses numbers after dates and figures
- L6:1469 excites
- L10:2461 excites
- L6:1563 excites
- L6:756 weakly suppresses
- L9:349 weakly suppresses the spot before a year
- L10:868 weakly suppresses
- L9:2160 weakly suppresses
- L10:2336 weakly excites
- L6:938 weakly excites
- L8:1244 weakly excites
- L6:679 weakly excites
- L4:209 weakly excites quantities right after 'a' in numbers
- L13:287 weakly suppresses
- L11:2532 weakly excites
- L9:2096 weakly suppresses 'the' or 'as' before a superlative role
- L13:1308 weakly excites
- L13:475 weakly suppresses
- L13:923 weakly suppresses word right before a number or year
- L6:1008 weakly excites 'of' after a measured quantity
Feeds later layers
- L15:368 excites
- L17:222 excites
- L16:256 excites number right after 'age of' or 'ages'
- L18:390 excites
- L19:58 suppresses
- L15:462 excites
- L16:211 excites
- L15:1909 excites blank slot before a number
- L17:46 excites
- L17:126 excites
- L15:448 excites
- L19:761 excites
- L16:253 weakly excites
- L21:557 weakly excites
- L26:154 weakly excites
- L21:175 weakly excites
- L16:372 weakly excites
- L17:124 weakly excites
- L15:828 weakly excites
- L22:9 weakly excites
- L15:1497 weakly excites
- L18:965 weakly excites
- L22:298 weakly excites
- L16:2137 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
- ITO
- 多久
- LEGRO
- парти
- 一世
- 丹
- Builders
- ogan
- 苹果
- izo
Suppresses
- льн
- tram
- draulic
- รณ
- embedded
- vest
- Ń
- venes
- رب
- ncia