slot before the second number in a range
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. 20 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them 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 spot right before the second number in a pair or range, such as the gap between '80' and '20' in '80% and 20%' or '80-90%'. Its negative side fires inside numbers, on a middle digit. Turning it up made the replies wander without a clear theme.
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.
-
wood, plaster, clay, or other substance to give it the colour and lustre of bronze. Dutch metal, an alloy of 80 percent copper and 20 percent zinc, is frequently used for bronzing.
-
-lash type injury. No imaging can identify a concussion and it is purely assessed on signs and symptoms that occur immediately or within a few days following.⏎80-90% of people with a concussion make a complete recovery within 1-
-
40% lean protein, 20% fat and 40% complex carbohydrates for optimum nutrition and satisfaction. One of the best things
-
ant/mood-stabilizing medication than the one you previously selected (other than lithium) that a psychiatrist might prescribe to treat bipolar disorders.⏎300 to 50
-
for the entire class.⏎PC: Windows XP and up⏎Mac: OS X (10.1) and up⏎Resolution: 800 x 600 or higher<|endoftext|>In response to the theme of 201
-
local wind shear was having its effect on the storm, and Hurricane Humberto’s winds had begun to diminish. By 1500 UTC (11:00 a.m), Humberto had weakened to tropical storm status, with maximum
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 0000000000000000000000000000000000000000000000
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
the early 1900s. The construction of the hotel began in 1901 and was completed in just two years, in 1903. The hotel was designed by Harrison Albright, an architect known for
-
mm) – this is one thousandth of a meter. In scientific notation, 1 millimeter is 1 – E3 m, which means 0.001 x 1
-
Consortium (CFSC) developed a three year C4E initiative (2007-2010) to pilot test the approach and tools articulated in the Practical Guidance Note
-
At the end of 1099 he visited Jerusalem along with Bohemund I of Antioch, but he returned to Edessa in January 1100. After Godfrey's death in 1100 he
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
- L17:1423 excites
- L19:2772 suppresses
- L16:2181 excites
- L17:1622 suppresses
- L17:1552 suppresses
- L18:1781 weakly excites
- L17:2082 weakly suppresses
- L19:1284 weakly suppresses digits inside years and decades
- L18:2858 weakly suppresses
- L19:616 weakly excites
- L13:2726 weakly excites
- L3:2926 weakly excites
- L19:2663 weakly excites
- L1:2404 weakly suppresses
- L17:107 weakly suppresses
- L13:2047 weakly suppresses
- L7:22 weakly suppresses
- L9:346 weakly suppresses
- L19:1851 weakly excites
- L9:42 weakly suppresses
- L8:2791 weakly suppresses
- L19:619 weakly suppresses
- L17:692 weakly suppresses
- L1:458 weakly suppresses
Feeds later layers
- L22:402 excites
- L23:775 suppresses
- L21:1732 suppresses
- L25:621 suppresses
- L23:2924 weakly suppresses
- L21:1053 weakly excites
- L25:1020 weakly suppresses
- L21:2918 weakly suppresses
- L24:797 weakly suppresses
- L22:1203 weakly suppresses
- L21:2428 weakly excites
- L21:972 weakly suppresses
- L25:492 weakly suppresses
- L22:2783 weakly suppresses
- L21:452 weakly excites
- L21:1182 weakly suppresses
- L23:1128 weakly suppresses
- L21:719 weakly excites
- L25:1662 weakly suppresses
- L23:1945 weakly suppresses
- L23:1411 weakly suppresses
- L26:751 weakly excites
- L24:992 weakly suppresses
- L22:1319 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
- igo
- LOSS
- >Type
- 蒟
- truth
- predic
- gettext
- FormatException
- (det
- 阵地
Suppresses
- rt
- iros
- v
- _v
- sz
- ствовать
- JD
- ulers
- 焉
- irable