the word 'whole'
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. 25 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 responds to the word "whole" in phrases like "whole number", "whole grains" or "whole egg". Turning it up made the model repeat the word "number" over and over, which fits the word it tends to predict next.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
to the nearest whole number.)<|endoftext|>Most of the sounds Sperm Whales produce are clicks, from
-
to reach their full potential by passively listening to the teacher and filling in worksheets they must engage in This is called a mixed number which is a mixture of a whole bbc bitesize to add the two fractions together print out the worksheet below from twink
-
for Americans recommends that at least half of all the grains you eat are whole grains. Chances are you eat lots of grains already. But are they whole
-
for your dataset may be different.⏎The stem-and-leaf plot allows us to see where data may be aggregated.⏎Use the app above to select a dataset with whole numbers. The app will output the dataset's stem-and-leaf plot once Submit
-
folks who have high blood pressure, even more evidence points to the benefits of the DASH diet (an eating plan that encourages a lot of fruits, vegetables, whole grains, and heart-healthy fats). Since these diseases and conditions are all too
-
one larger whole.⏎As it becomes more difficult to shrink transistors in size, and impr
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: the best way to fit together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together together
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
ucius’ philosophy revolved around two concepts: the nobleman and the establishment of a well-ordered society. The nobleman (君子, pinyin: jū
-
no leader since the priest or priestess read the will of deities.⏎Leadership, Odin, messenger birds, Lugh Long-arm, Odin, Tyr, Amled saga, neck-ring, Cernunnos, the Stag, Trinity
-
Perseids meteor shower.⏎There are many meteor showers during the year, but the Perseids in August, the Leonids in November, and the Geminids in December are predict
-
largest hang like many-limbed chandeliers, turning slowly in the current. When I get closer, I can see the individual polyps, the anemone-like creatures that make up each fragmented colony. Nestled in the cren
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
- L18:242 weakly suppresses
- L11:1091 weakly excites
- L23:270 weakly excites
- L9:1795 weakly suppresses
- L2:2863 weakly suppresses
- L24:105 weakly suppresses
- L16:819 weakly excites
- L1:987 weakly excites
- L13:2877 weakly suppresses
- L3:1939 weakly suppresses
- L17:1061 weakly suppresses
- L3:62 weakly suppresses
- L20:183 weakly excites
- L2:343 weakly suppresses
- L3:508 weakly excites the final zero of decades like '1980s'
- L0:2571 weakly suppresses
- L10:933 weakly suppresses
- L5:2466 weakly excites
- L4:1323 weakly suppresses
- L16:11 weakly suppresses
- L22:2003 weakly suppresses
- L10:190 weakly suppresses
- L11:2637 weakly suppresses
- L7:846 weakly excites
Feeds later layers
- L26:861 excites
- L27:974 excites
- L27:1479 weakly excites
- L26:2757 weakly excites
- L27:2145 weakly suppresses
- L26:880 weakly suppresses
- L27:572 weakly suppresses
- L26:2628 weakly excites
- L27:1568 weakly excites
- L26:1618 weakly excites
- L27:886 weakly excites
- L27:2484 weakly suppresses
- L27:1389 weakly suppresses
- L26:748 weakly suppresses
- L27:1812 weakly suppresses
- L27:229 weakly suppresses
- L27:1631 weakly suppresses
- L27:2910 weakly suppresses
- L27:2795 weakly suppresses
- L26:969 weakly excites
- L27:1944 weakly excites
- L27:1754 weakly suppresses
- L26:1520 weakly excites
- L27:634 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
- 数
- number
- number
- 數
- numbers
- (number
- số
- Number
- -number
- num
Suppresses
- iggs
- @Json
- Frederick
- EDURE
- opause
- chaft
- Inflate
- QGraphics
- iedad
- FITNESS