"wide" after a measurement
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. 4 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 lights up on the word "wide" when it describes how broad something is, such as "25 feet wide" or "a wide range". Turning it up made the model write unrelated text, so the steering did not show this.
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.
-
sand or loam soil. Mature specimens typically range from 20 to 30 feet high and up to 25 feet wide. Eastern redbud has a uniform, vase-shaped habit with multiple trunks
-
id lowland coastal region, the central high sierra of the Andes, and the dense forest of the Amazon. A wide range of important mineral resources are found in the mountainous and coastal areas, and
-
Hawaii and California. We discovered that the conventional method of measuring ocean plastic, using nets of less than a meter (3 ft) wide, was inaccurate because it seriously underestimated the total amount of plastic. The reason for
-
! There’s more…⏎How Science Works has been adapted as an online course for middle- and high-school science educators to “broaden their own knowledge and understanding and to use with students, the course weaves together
-
–22" tall⏎Variety Information: Large, dark green oval leaves with serrated edges up to 10" long, and 8" wide⏎Attributes: Good for Containers⏎When to Sow Outside: RECOMMENDED
-
actually glass set in a steel frame set in the wooden door a method that was at that time also used in palaces in Budapest and Vienna.⏎15 meter wide at the front and 65 meters deep it has 500 square
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: (6.0.2.1.1.1.1) I think about the number How to pick the best I am really confused to the fact Which is the best time to get together to think about
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
is the main ingredient in your diet.⏎Food is food insofar as our tastes are concerned. But, when it comes to our stomachs, and, by extension, our whole body is involved, food is more than just food. It’s
-
your tree is essential to its health and survival for the first few years after it’s been planted. Watering slowly and deeply is better for your tree’s over-all health then fast and frequent waterings. Watering before 10 a.m
-
marshes and clappers in salt. The two birds overlap in brackish water marshes, and they do hybridize. Clapper rails are darker over all, with a mix of grays and deep browns that kings don’t have
-
If people mimic or follow someone else’s<|endoftext|>|Image Credit :|⏎Delhi, officially the capital territory of India is a collection of a number of cities spread across the metropolitan region and has been continuously inhabited since the 6th century BC.
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
- L2:2273 suppresses
- L3:272 weakly excites
- L3:2925 weakly suppresses
- L2:2400 weakly suppresses
- L1:1464 weakly excites
- L3:573 weakly excites
- L3:691 weakly excites
- L3:2037 weakly suppresses
- L3:2482 weakly excites
- L2:111 weakly suppresses
- L3:2508 weakly suppresses
- L2:2328 weakly suppresses
- L3:122 weakly excites
- L2:1036 weakly excites
- L3:102 weakly suppresses
- L3:464 weakly excites
- L2:1827 weakly excites
- L0:3016 weakly excites
- L3:187 weakly excites
- L2:2736 weakly suppresses
- L3:1525 weakly excites
- L2:2725 weakly suppresses
- L0:1224 weakly suppresses
- L2:75 weakly suppresses
Feeds later layers
- L24:2779 excites
- L8:1466 excites
- L7:2489 excites
- L22:2961 excites
- L27:1610 weakly excites
- L26:1154 weakly excites
- L5:1087 weakly excites
- L18:374 weakly excites
- L15:927 weakly excites
- L19:323 weakly suppresses
- L19:1340 weakly excites
- L5:1806 weakly excites
- L7:2741 weakly suppresses
- L8:1274 weakly excites
- L18:2892 weakly excites
- L10:1033 weakly excites
- L22:1838 weakly suppresses
- L22:2453 weakly suppresses
- L5:1247 weakly excites
- L18:919 weakly suppresses
- L20:470 weakly suppresses
- L8:2986 weakly suppresses
- L26:2129 weakly suppresses
- L18:533 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
- →width
- (width
- wider
- width
- .width
- WIDTH
- """⏎⏎⏎
- width
- """
- widen
Suppresses
- ride
- .GetObject
- ign
- 干事
- ivol
- 红星
- Ronnie
- volumes
- rides
- .ISupportInitialize