line break after a title or heading
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. 0 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 line break that comes right after a short title or heading, such as a page name or a section label, before the body text starts. Turning it up did not produce a clear, related change in the output.
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.
-
th Edition⏎This is Herb's step-by-step introduction to Java, updated for⏎Java SE
-
Suggested Learning Target⏎- I can explain the difference between a.m. and p.m.⏎- I
-
atta Facts⏎Mohammad Hatta (1902-1980), one
-
New York City⏎Suggested Time Allowance: 1 hour⏎1. Participate in a
-
aka Peace Pagoda⏎In 1993, the foundations to the Dhamma Talaka
-
Amazon Web Services⏎|Type of site||Web service, cloud computing|⏎Amazon Web Services (AWS
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: " (1) is the most important thing in the world. It's the most important thing in the world. It's the most important thing in the world. It's the most important thing in the world. It's the most important
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
fish caused by acid rain.⏎Brodde Almer, Institute of Freshwater Research, Drottningholm investigated in the summer of 1970
-
symptoms.⏎Dosage information relates to the recommended dosage and frequency of drugs a patient should consume to treat a specific medical condition<|endoftext|>A species on the brink – Freshwater Pearl Mussel⏎They can live for over a century, they have one of
-
is the only major North Carolina estuary having a direct connection to the sea. Short-term flow throughout most of its length is dominated by ocean tides. Freshwater entering the major estuaries is, where not contaminated, of acceptable quality for
-
including the world’s largest mammals, the blue whales. The ocean’s breadth and depth remain the most extensive, unexplored habit.⏎Activity 2 – Freshwater Habitats⏎Freshwater Habitats cover less than 0.01 %
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
None: this is the first layer.
Feeds later layers
- L2:1574 weakly excites
- L3:2053 weakly suppresses
- L13:1774 weakly suppresses
- L16:2478 weakly suppresses
- L12:1279 weakly suppresses
- L17:2863 weakly suppresses
- L5:2045 weakly excites
- L15:1359 weakly suppresses
- L16:232 weakly excites
- L5:1491 weakly excites tight, narrow, constricted (squeezed words)
- L6:3032 weakly excites
- L11:1935 weakly excites
- L12:1053 weakly excites
- L25:1022 weakly excites
- L8:2320 weakly suppresses
- L9:612 weakly suppresses
- L18:1916 weakly excites
- L3:2753 weakly suppresses
- L22:22 weakly excites
- L11:2043 weakly suppresses
- L27:44 weakly excites
- L1:2754 weakly excites
- L6:1713 weakly suppresses
- L13:2040 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
- actors
- рам
- reo
- ◄
- ระ
- roduction
- ро
- 론
- rockets
- рам
Suppresses
- 藜
- insic
- orarily
- handed
- opia
- lich
- ilage
- 外科
- 合影
- isp