zeros after a decimal point
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. 24 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 reacts to the zero in small decimal numbers like 0.01, 0.2 or 0.003, the zeros that come right after the decimal point. When it was turned up, the model often began writing long strings of zeros.
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 it being the basis of every single financial transaction.⏎In today’s American coinage, there are six circulating denominations. These are the penny ($0.01
-
also sometimes called step size) to determine the next point. For example, if the gradient magnitude is 2.5 and the learning rate is 0.01, then the gradient descent algorithm will pick the next point 0.0
-
here to learn more about weed control and weed and feed.⏎Clay soil is comprised of small, tightly packed particles with a diameter less than 0.002mm. If you examine the structure of these particles, you’ll find it
-
will make a microscope slide, or thin section, of their rock of interest. A thin section has a very thin slice of rock, usually about 0.03 mm thick, attached to a glass slide with an epoxy resin. Thin sections
-
by the Modelica Association.⏎This example demonstrates different noise distributions methods that can be selected for a Noise block. Both noise blocks use samplePeriod = 0.02 s, y_min=-1, y_max=3, and have identical
-
breadth and depth remain the most extensive, unexplored habit.⏎Activity 2 – Freshwater Habitats⏎Freshwater Habitats cover less than 0.01 % of the earth’s
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 state of a game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game game
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the National Hurricane Center can keep tabs on them.⏎Once a cyclone’s winds kick up to 39 miles per hour and sustain that speed for 10 minutes, it becomes a tropical storm and the NHC gives it a name (
-
2, 144, 180, and 200, resulting in stepping angles of 30, 15, 5, 2.5, 2, and 1.8° per
-
a straight line. The reason the path is all bendy is because Swift itself is in motion! Swift orbits the Earth, circling us once every 90 minutes or so. As it moves around us, its viewpoint is changing, and
-
- Hernias: Hernia is the popular term for a series of different diagnoses, all identified by outpouchings on the ventral aspect of the abdominal wall of pigs. Hernias can be painful and limit
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
- L23:2500 excites
- L19:843 suppresses
- L21:3057 suppresses
- L20:626 excites
- L4:498 weakly suppresses
- L20:2749 weakly suppresses
- L23:1324 weakly suppresses
- L10:1052 weakly suppresses
- L3:1124 weakly suppresses
- L10:726 weakly suppresses
- L9:977 weakly suppresses the last digit of a decade year
- L20:1504 weakly excites
- L16:184 weakly suppresses
- L7:2244 weakly suppresses
- L9:344 weakly suppresses
- L2:976 weakly suppresses
- L5:935 weakly suppresses
- L10:934 weakly suppresses
- L12:1587 weakly excites
- L6:2792 weakly excites
- L21:203 weakly excites
- L0:997 weakly suppresses
- L13:2294 weakly suppresses
- L1:3053 weakly excites
Feeds later layers
- L25:473 excites zeros inside large round numbers
- L26:413 excites
- L27:1153 excites
- L27:862 excites
- L27:1196 excites
- L27:153 suppresses the spot just before a historical number
- L27:396 excites
- L27:530 suppresses
- L25:182 weakly excites
- L25:92 weakly excites
- L26:121 weakly excites
- L26:737 weakly excites
- L27:1111 weakly excites
- L26:682 weakly suppresses
- L25:212 weakly excites
- L25:543 weakly suppresses
- L25:1308 weakly excites
- L25:324 weakly suppresses
- L26:233 weakly suppresses
- L25:2243 weakly suppresses
- L25:2022 weakly excites
- L27:628 weakly suppresses
- L25:1895 weakly excites first digit of a year like 2020
- L27:1029 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
- 0
- 零
- zero
- zero
- htag
- ₀
- 0
- Zero
- heets
- zept
Suppresses
- 驷
- ystate
- <State
- fine
- affen
- ucid
- 该游戏
- 衫
- loses
- 輩