Neuron Explorer

Qwen3-0.6B

colon or equals sign before a definition

A neuron in layer 10 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

The neuron fires on the colon or equals sign that comes right before a definition, label or value, such as the = in a formula or the : after a heading word. It was mostly quiet on ordinary sentences. Turning it up made the model's text wander into unrelated first-person sentences, which doesn't clearly relate to the trigger.

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

    ACC = E/V × Re + D/V × Rd × (1 - Tc)⏎Re = cost of equity⏎Rd = cost of debt⏎E = market value of the firm's equity⏎D = market value of the firm's

  2. #2 peak

    1999; Grevemeyer et al. 2010). The rectangle shows the position of Fig. 1b.⏎b :Map of eruption centers and main fissures of the 1995 eruption

  3. #3 peak

    information about visited websites and started applications. It is also measuring time in front of the monitor.<|endoftext|>DRDO Placement Paper (EMT & Microprocessor)⏎Paper: DRDO Placement Paper (EMT & Microprocessor)⏎EMT and microprocessor

  4. #4 peak

    .5, 2, and 1.8° per step.⏎Step angle = 360°/No of steps per revolution⏎Where m = Number of Phase winding on the

  5. #5 peak

    console in front of the furnace), etc., The electrical part of the large medium frequency furnace also includes a crucible leak alarm system.⏎Furnace body part: medium and small medium frequency furnaces are equipped with two furnace bodies. One is

  6. #6 peak

    -fetching. Relevant Computer Science topics such as models of computation and formal grammars are addressed, and their practical value explained.⏎Share Low-Level Programming<|endoftext|>New Delhi: For possibly the first

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: and I'm going to think about it. and I'm going to think about it. and I'm going to think about it. and I'm going to think about it. and I'm going to think about it. and I'm

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

    of life.⏎If anything, this emphasizes the importance of taking care of your physical and mental health through college and

  2. #2 peak

    is not simply whether road pricing is regressive, or even if it will burden the poor. The relevant question is whether road pricing will burden the

  3. #3 peak

    were gone."⏎Join to answer this question⏎Join a community of thousands of dedicated teachers and students.Join eNotes<|endoftext|>Is there something else you would like to know?⏎Use the alphabetical glossary listing to

  4. #4 peak

    the message, his answer is merely to turn up the volume and repeat the message as a tautology: US policy is democratic because America is free. One suspects that if there was more truth to the assertion it would not have to be repeated

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. L9:396 excites colon after a metadata label
  2. L4:24 excites
  3. L9:319 excites equals signs in maths and equations
  4. L3:77 excites
  5. L4:193 excites
  6. L6:1609 suppresses
  7. L8:309 weakly excites
  8. L8:137 weakly suppresses
  9. L0:2259 weakly suppresses
  10. L3:76 weakly suppresses
  11. L5:1976 weakly excites
  12. L6:1752 weakly excites
  13. L7:3018 weakly suppresses
  14. L4:113 weakly suppresses
  15. L3:294 weakly excites
  16. L6:2843 weakly suppresses
  17. L2:357 weakly excites
  18. L6:2237 weakly excites
  19. L4:701 weakly suppresses
  20. L6:482 weakly suppresses
  21. L4:302 weakly excites blank slot before a number
  22. L0:877 weakly excites
  23. L4:2690 weakly excites
  24. L4:2769 weakly suppresses

Feeds later layers

  1. L13:1343 excites
  2. L16:2177 excites
  3. L12:1150 excites
  4. L15:1041 excites
  5. L16:1142 excites
  6. L17:1066 excites
  7. L16:466 excites 'is' after a long subject phrase
  8. L12:2562 excites
  9. L14:1769 excites
  10. L12:850 excites
  11. L12:2171 excites
  12. L13:2292 excites opening parenthesis for a gloss or abbreviation
  13. L12:1212 suppresses
  14. L13:652 excites
  15. L13:371 weakly excites
  16. L13:948 weakly excites
  17. L14:827 weakly excites open bracket after a value or citation
  18. L12:792 weakly suppresses
  19. L11:502 weakly excites
  20. L11:567 weakly suppresses
  21. L14:995 weakly excites
  22. L24:154 weakly suppresses
  23. L11:162 weakly excites
  24. L16:551 weakly suppresses

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.

2 values between -2.10 and -1.961 values between -1.54 and -1.402 values between -1.40 and -1.266 values between -1.12 and -0.986 values between -0.98 and -0.8411 values between -0.84 and -0.7015 values between -0.70 and -0.5654 values between -0.56 and -0.43134 values between -0.43 and -0.29539 values between -0.29 and -0.151482 values between -0.15 and -0.011293 values between -0.01 and 0.13243 values between 0.13 and 0.2756 values between 0.27 and 0.4125 values between 0.41 and 0.558 values between 0.55 and 0.696 values between 0.69 and 0.8310 values between 0.83 and 0.973 values between 0.97 and 1.112 values between 1.11 and 1.254 values between 1.25 and 1.391 values between 1.81 and 1.952 values between 1.95 and 2.081 values between 2.22 and 2.361 values between 2.64 and 2.781 values between 3.34 and 3.48offnegativepositiveoften

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. offsetof
  2. ieux
  3. immers
  4. istani
  5. oubles
  6. 不错
  7. 每人
  8. ional
  9. CTOR
  10. (HttpStatus

Suppresses

  1. Disposed
  2. 蜂
  3. 鹩
  4. @(
  5. 游戏当中
  6. ificant
  7. себя
  8. niece
  9. ORTH
  10. UW