Neuron Explorer

Qwen3-0.6B

weighing, counting and measuring

A neuron in layer 5 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on words about measuring things: weighing gold, counting stars or steps, using a balance, measuring something. When turned up, the model's text drifted toward temperatures and thermometer readings, a loose match to the same 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 activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    , go to the place marked out by the Lord your God for himself;⏎Gold cannot be exchanged for it,and silver cannot be weighed out for its price.⏎Gold and glass do not compare with it,and articles of

  2. #2 peak

    “Look now toward heaven, and count the stars if you are able to number them…. So shall your descendants be” (

  3. #3 peak

    stration: This could be a useful introduction to the idea of density of a liquid.⏎Demonstration: This method uses an equal-arm balance rather than a domestic spring balance.⏎Class practical: A simple method for comparing the density of liquids.⏎There is

  4. #4 peak

    people have resumed their morning and evening walks in the neighbourhood parks/areas. It’s the same walk with the same shoes with cushioned soles.⏎You count your steps on the app in

  5. #5 peak

    20th century. Some historians consider the memory of the Shoah to be a new civil religion of the western world, a negative yardstick by which to measure the legitimacy of our democratic institutions.⏎Everyone agrees that the Shoah is a major

  6. #6 peak

    standard filmmakers today still are struggling to measure up to,” he said.⏎Kaufman said the Disney film is based,

Steering

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.

turned down a lot
Prompt

Here is something I have been thinking about lately: a little problem, a question. I need to find a way to represent this in a better format. Let me think it up, but I can't get the image to show up. Let me think it's better. Let me think

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    quality of the composting process. It is recommended to always add carbon materials in shredded form and food scrapes in small pieces. Small particle size increases the surface area and helps the microorganisms to decompose organic waste quickly.⏎The two main stages

  2. #2 peak

    02 and 2003. Sediment delivery during each pulse was highly variable (11,300-43,800 metric tons), and was greatest during rising limbs of Mississippi River flood events. Overland

  3. #3 peak

    medium under controlled environmental conditions.⏎This propagation method produces plants of uniform genetic makeup, increases the rate of plant production and has been used for the elimination of diseases in certain plants.⏎T

  4. #4 peak

    surface friction and wear between interacting surfaces, to dissipate heat, prevent corrosion and transfer of power. In general, a good lubricant should have a high viscosity index, high flash and fire points (higher than the operating temperature of the machine),

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L3:2935 excites
  2. L4:1939 suppresses
  3. L4:2662 excites
  4. L3:2405 excites
  5. L3:2615 excites
  6. L0:249 weakly excites ordinal suffix "nd" after numbers
  7. L4:3005 weakly excites
  8. L3:2872 weakly excites
  9. L3:1973 weakly suppresses
  10. L4:1024 weakly suppresses
  11. L4:1862 weakly suppresses
  12. L3:2282 weakly excites
  13. L2:2143 weakly excites
  14. L3:1708 weakly suppresses
  15. L4:2251 weakly excites
  16. L4:3053 weakly suppresses
  17. L3:1456 weakly excites
  18. L0:1851 weakly suppresses
  19. L2:1148 weakly excites
  20. L4:1399 weakly excites
  21. L4:25 weakly excites
  22. L3:2634 weakly suppresses
  23. L2:2989 weakly excites
  24. L4:926 weakly excites

Feeds later layers

  1. L10:127 excites
  2. L10:1538 excites
  3. L9:712 excites
  4. L8:1589 excites
  5. L10:294 excites
  6. L6:175 suppresses
  7. L7:91 suppresses
  8. L17:1735 excites
  9. L23:549 excites
  10. L9:660 excites
  11. L19:1603 excites
  12. L8:1466 excites
  13. L11:95 suppresses
  14. L11:2223 excites
  15. L21:1450 weakly excites
  16. L10:2214 weakly excites
  17. L20:2792 weakly excites
  18. L7:2042 weakly excites
  19. L11:781 weakly excites
  20. L10:229 weakly excites
  21. L25:1250 weakly excites
  22. L9:806 weakly excites
  23. L10:1967 weakly excites
  24. L9:391 weakly excites

Activation histogram

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.

1 values between -1.45 and -1.223 values between -0.76 and -0.5363 values between -0.53 and -0.311152 values between -0.31 and -0.082457 values between -0.08 and 0.15199 values between 0.15 and 0.3814 values between 0.38 and 0.615 values between 0.61 and 0.845 values between 0.84 and 1.062 values between 1.06 and 1.292 values between 1.29 and 1.521 values between 1.52 and 1.752 values between 1.75 and 1.981 values between 1.98 and 2.211 values between 7.46 and 7.68offnegativepositiveoften

Logit lens

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 layers instead.

Promotes

  1. 尺
  2. second
  3. peso
  4. everything
  5. tres
  6. scale
  7. 测
  8. то
  9. measurement
  10. second

Suppresses

  1. .']
  2. ображен
  3. 一角
  4. ellant
  5. ogenesis
  6. ==='
  7. istique
  8. formation
  9. 阙
  10. unpl