digits inside years and numbers
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. 26 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 number characters, especially the third digit of years like 2020 or 2023 in dates. When it was turned up, the model started writing number words like two, four and five, and when turned down it repeated the digit 3.
Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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.
-
a potential lunar lava tube entrance – could offer insightful data concerning the moon’s geological history and could hint<|endoftext|>(Updated on June 22, 2020)⏎The mass air flow (MAF) sensor is an important component which
-
appearance of troops at Pine Ridge, he advocated for peace with the European Americans⏎Date de dernière mise à jour : 05/07/2021⏎Ajouter un commentaire<|endoftext|>By DANIEL SUDDEATH⏎NEW
-
, osteoporosis, meniscus, ligament injuries, muscle problems.⏎Günceleme: 07/01/2023 14:52<|endoftext|>Sustainability and Renewable Merino⏎W
-
supports other economic activity, provides an essential product and is poised for long-term growth⏎By Rob Roach, ATB Economics 22 February 202
-
part of the city nearest<|endoftext|>Measurement is an important part of mathematics learning.Which statement best about young children learning to measure? January 1, 2023 by admin Measurement can only be done when children can use standardized tools. Measurement
-
? Scholarship application essay example this column presents those 10 problems and provides solutions to each. how do you know? Apr 29, 2020 · problem-solving strategies
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 no idea what I do for ound I live." 3 ound I live ound ound I live ound ound ound
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
to the ip address. The number of port in computer networking is expressed in 16 bits, which means that there is a maximum of 65,536 ports on a computer. Assign the call coverage path for voice ports b
-
Fig. 1. Electron microscopy of the viral-like particles generated by the Phoenix provirus transfected into human 293T cells. The prominent spikes of the Env protein are visible under hight
-
at least twice a week, but you can also get omegarich fish oils as dietary supplements, though they may not have the DHA and EPA omega-3s specifically found in fatty fish. Oatmeal is high in soluble fiber
-
such as broccoli, cabbage, and carrots. Heart-healthy eating includes fat-free or low-fat dairy products, such as, skim milk, fish high in omega-3 fatty acids, such as salmon, tuna, trout, about twice a week
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
- L25:1040 strongly suppresses digit 1 inside dates and times
- L22:656 strongly excites
- L25:1749 strongly suppresses
- L21:2789 strongly suppresses
- L22:253 strongly excites
- L23:1625 strongly excites
- L25:230 strongly suppresses
- L23:1178 strongly excites
- L23:358 strongly excites
- L21:731 strongly excites
- L24:2522 strongly suppresses
- L23:284 suppresses
- L24:2616 suppresses
- L20:1011 suppresses
- L23:450 excites unclear: raccoon name fragments
- L25:1020 suppresses
- L24:942 excites
- L21:2036 suppresses
- L21:563 excites
- L24:1949 excites
- L21:2064 excites
- L25:2116 excites
- L21:1211 suppresses
- L22:1016 suppresses
Feeds later layers
- L27:364 strongly suppresses
- L27:555 suppresses
- L27:678 suppresses
- L27:992 suppresses
- L27:391 excites
- L27:352 excites
- L27:205 suppresses
- L27:435 weakly excites
- L27:572 weakly excites
- L27:496 weakly excites
- L27:448 weakly excites
- L27:628 weakly excites
- L27:1440 weakly excites
- L27:328 weakly excites
- L27:189 weakly suppresses
- L27:763 weakly excites
- L27:2373 weakly suppresses
- L27:893 weakly excites
- L27:2679 weakly suppresses
- L27:1232 weakly excites
- L27:2290 weakly excites
- L27:978 weakly suppresses
- L27:1100 weakly excites
- L27:2590 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
- 四种
- 四项
- 两张
- 五个
- 四个
- 八个
- four
- fourteen
- Four
- Four
Suppresses
- 3
- 3
- three
- -three
- الثالث
- ³
- 三百
- _three
- trois
- drei