"of" after a measurement noun
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. 1 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 word "of" when it follows a measurement word such as "rate", "temperature", "resolution" or "capacity", as in "a flow rate of 16 l/min". Turning it up did not make the model's writing clearly about measurements.
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.
-
argon and 5% CO 2 gas mixture at a flow rate of 16 l/min was used for shielding throughout the welding.⏎Some of fab
-
year in Mount Surprise is 73.0°F (22.8°C). The warmest month, on average, is November with an average temperature of 81.0°F (27.2°C). The coolest month
-
asa said.⏎Colors of Mercury⏎The project is part of a global color map that covers more than 99% of Mercury’s surface with an average resolution of about 1 kilometer per pixel. The view here is what an observer would
-
the image has a resolution of about 70 meters (230 feet) per pixel. This image
-
what you want your error of tolerance to be. For example, you may be happy with a confidence level of 95 percent (giving a margin error of 0.05), or you may require a tighter accuracy of a
-
taken in visible light with the Cassini spacecraft narrow-angle camera on Jan. 22, 2008. The view was obtained at a distance of approximately 1.7 million kilometers (1.1 million miles) from Prometheus
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: I have been thinking about lately, and I have been thinking about lately. How do you think about lately? How do you think about lately? How do you think about lately? The question is about how 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.
-
global event day and raise the profile of cancer in people’s minds and in the world’s media
-
be 99,800 unaffected pregnancies and 200 pregnancies with Down syndrome. The table below compares the results of the most commonly used biochemical screening test (
-
of heat.⏎Made of shredded newspaper and a fire retardant chemical (borate), the paper is broken down into cellular fibers.⏎WHAT
-
. Election day was largely peaceful and approximately 81% of voters cast ballots, one of
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
- L0:1224 suppresses
- L0:178 suppresses
- L0:1301 suppresses
- L0:93 weakly suppresses
- L0:1211 weakly excites
- L0:2296 weakly excites
- L0:250 weakly suppresses
- L0:2799 weakly suppresses
- L0:2916 weakly excites
- L0:2751 weakly suppresses
- L0:638 weakly suppresses
- L0:1950 weakly excites
- L0:1672 weakly suppresses
- L0:265 weakly suppresses
- L0:788 weakly suppresses
- L0:267 weakly excites
- L0:1715 weakly excites
- L0:1654 weakly excites
- L0:2828 weakly suppresses
- L0:1527 weakly excites
- L0:1826 weakly suppresses
- L0:2299 weakly excites
- L0:3031 weakly suppresses
- L0:1583 weakly excites
Feeds later layers
- L3:312 excites
- L12:1570 excites
- L18:585 excites
- L11:105 weakly suppresses
- L10:1034 weakly excites
- L18:2737 weakly suppresses
- L4:106 weakly excites
- L18:2991 weakly excites
- L21:175 weakly excites
- L19:1219 weakly excites
- L24:722 weakly excites 'about' or 'approximately' before a number
- L14:840 weakly excites
- L3:540 weakly excites
- L8:1500 weakly excites
- L24:1394 weakly excites
- L19:1947 weakly excites
- L12:1467 weakly excites numbers after dates and figures
- L22:1497 weakly suppresses
- L10:868 weakly excites
- L6:1214 weakly suppresses
- L6:231 weakly excites
- L15:828 weakly excites
- L11:2896 weakly suppresses
- L19:1888 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
- amento
- 不低于
- abilities
- 第一百
- 旬
- 睡
- atisf
- ativity
- lickr
- riott
Suppresses
- ichern
- 身上
- eling
- Hast
- 手上
- ˣ
- 담
- helm
- oval
- whole