"of" after a measure like inches or teaspoon
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. 12 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
Lights up on "of" right after a measurement or portion word, such as "an inch of", "a teaspoon of" or "several inches of". Turning it up made the output wander into repeated fragments that did not match the prompts.
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.
-
later blooms, sow them after the last frost date.⏎- If you are planting seeds, sow them directly in the ground, covering them with about ¼ inch of soil. Space them 8 to 18 inches apart, depending on the
-
a class in Java? Local variables − Variables defined inside methods, constructors or blocks are called local variables. Classes define the behavior for objects of its type. Instances of the class Class represent classes and interfaces in a running Java application. If we do
-
Oxide 0.4% Sulphur Dioxide 0.4%; Phosphorous Oxide 0.3%.⏎Take about one teaspoon of food-grade diatomaceous earth with liquid once
-
Vietnamese. The division of Cham into Western and Phan Rang Cham immediately followed the Vietnamese overthrow of the last Cham polity. Each uses a distinct variety of the script, although the former are mostly Muslim and now prefer to use the Arabic alphabet
-
sitting in muddy soil or in an area where water is pooled on top of the soil.⏎Mulch around the base of outdoor gardenia shrubs with several inches of organic material
-
, but the name Mohs chemosurgery continues to be associated with the procedure now correctly termed Mohs Micrographic Surgery.⏎After the removal of the visible portion of the tumor by excision or curettage (debulking), there are
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 need to do a question for a report. You can find the link to the article in the link. But since I'm not sure about the answer, I just need to make sure. So, the user wants to put in a
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
concerned and demonstrates their free, prior and informed consent;⏎- R.5: A large number of local manifestations of the element are inscribed on the National List of Intangible Cultural Heritage administered by the Department of Intangible Cultural Heritage of the Ministry
-
one-to-one functions. DEFINITION A function is called a one-to-one function if it never takes on the same value twice; that is, If a horizontal line intersects the graph of in more than one point, then we see from Figure
-
(at least to Physicists!) experiment, Galileo went to the top of the leaning tower of Pisa and dropped a wooden ball and a lead ball, both the same size
-
Red List of Threatened Species.⏎The ocean is the bloodstream of mother earth. The Artery
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
- L10:1395 excites
- L9:1428 excites
- L10:469 suppresses 'of' after a noun phrase
- L11:579 suppresses
- L9:463 excites
- L6:3070 excites
- L11:153 suppresses
- L8:1003 suppresses "the" or "of" before a noun phrase
- L5:453 suppresses
- L8:502 excites
- L8:892 excites
- L10:1364 weakly excites
- L11:601 weakly suppresses
- L8:1212 weakly excites
- L6:1557 weakly excites
- L5:1721 weakly excites
- L6:2217 weakly suppresses
- L8:526 weakly suppresses
- L9:592 weakly excites
- L4:1686 weakly excites
- L5:363 weakly excites
- L1:664 weakly suppresses
- L8:1755 weakly suppresses
- L9:393 weakly suppresses
Feeds later layers
- L16:2530 excites
- L16:1216 suppresses
- L16:961 excites
- L14:2107 excites
- L14:1971 excites
- L19:1503 excites
- L15:632 excites
- L13:1781 excites
- L13:486 excites
- L15:1028 excites
- L16:1077 excites
- L20:956 excites
- L17:663 excites
- L16:719 excites
- L16:2612 excites
- L13:1997 excites
- L15:404 suppresses
- L14:1508 excites
- L13:1906 excites
- L14:1707 excites
- L14:1901 excites
- L17:1241 excites
- L16:408 excites 'of' after a quantity or container noun
- L13:602 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
- 谋
- tactic
- pter
- 정도
- quent
- ami
- assumption
- ولد
- of
- CRM
Suppresses
- isée
- مل
- 茬
- neau
- iddle
- �
- 呛
- isseur
- itore
- 让她