Neuron Explorer

Qwen3-0.6B

opening bracket before an acronym

A neuron in layer 7 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

This neuron fires on an opening parenthesis that comes right after the full name of an organisation and before its short form, such as 'World Health Organization (WHO)'. The negative side is weak and comes from full stops and line breaks in lists. Turning it up made the model produce repetitive or off-topic text.

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

    remedy given by the trial judge was improper. The process will now be remitted back to the planning stage that was previous to the Peel Watershed Planning Commission’s (the “Commission”) Final Recommended Report (“Final Recommended Plan”), instead of being rem

  2. #2 peak

    on the calyx and stem end of the fruit.<|endoftext|>The American Lung Association's (ALA's) Not-On-Tobacco (N-O-T) web site asserts

  3. #3 peak

    square centimeter.⏎PG&E's electric smart meters and Alameda Municipal Power's (AMP) electric smart meters have the same level of radio frequency waves at 9

  4. #4 peak

    you to eat more produce: The government has gotten on the bandwagon. According to the U.S. Department of Agriculture’s (USDA) and Department of Health and Human Services’ (HHS) Dietary

  5. #5 peak

    residing in communities.⏎Two national cross-sectional surveys using a two-stage cluster sampling design were conducted among the Lebanese and Syrian refugee adults.⏎We used the World Health Organization (WHO) STEPwise approach through questionnaire assessment and physical and biochemical measurements. All reported

  6. #6 peak

    , December 2, 2012.⏎The contest is a partnership with The Planetary Society in Pasadena, California; the Massachusetts Institute of Technology’s (MIT) Lincoln Laboratory in Lexington, Massachusetts; and the University of Arizona in Tucson

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: get a sense of how far you go in the future, you go to the front, you go to the back...you can't stop it...now, you can't stop it...then you can't go back to the front...

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

    It’s not failure if you enjoyed the process.⏎28. I don’t believe in failure. It is not failure if you enjoyed the process.⏎29. It is on our failures that we base a new and different and better success.⏎

  2. #2 peak

    Prev Med Rep⏎Authors: Altman EA, Lee KL, Hecht CA, Hampton KE, Moreno G, Patel AI⏎Examining Policies Providing Access to Free Water in Schools and the Health Impact of an

  3. #3 peak

    spring if your heart is filled with past failures.⏎20. Don’t think of it as failure. Think of it as time-released success.⏎21. There are no failures. Just experiences and your reactions to them.⏎22.

  4. #4 peak

    )|⏎|Drives on the||right|⏎|1 Latvia is de jure continuous with its declaration November 18,

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. L3:21 excites
  2. L5:114 suppresses
  3. L4:260 excites
  4. L1:1221 weakly excites
  5. L2:1231 weakly suppresses
  6. L4:1501 weakly suppresses
  7. L4:382 weakly suppresses
  8. L3:42 weakly excites
  9. L5:1403 weakly suppresses
  10. L6:82 weakly excites
  11. L4:1750 weakly excites
  12. L2:3030 weakly excites
  13. L3:1099 weakly suppresses
  14. L3:1622 weakly suppresses
  15. L4:6 weakly suppresses
  16. L4:887 weakly excites
  17. L3:2278 weakly excites
  18. L0:639 weakly excites
  19. L4:1242 weakly excites
  20. L4:1963 weakly excites
  21. L0:2521 weakly suppresses
  22. L2:551 weakly excites
  23. L4:2734 weakly excites
  24. L2:273 weakly excites

Feeds later layers

  1. L13:236 excites
  2. L10:330 excites
  3. L19:527 weakly excites
  4. L15:632 weakly suppresses
  5. L21:227 weakly suppresses
  6. L14:451 weakly excites
  7. L21:1956 weakly suppresses
  8. L10:1543 weakly excites
  9. L25:626 weakly excites
  10. L8:337 weakly excites
  11. L15:431 weakly suppresses
  12. L21:161 weakly suppresses
  13. L17:332 weakly suppresses
  14. L15:612 weakly suppresses
  15. L11:109 weakly excites dash at start of list item
  16. L9:705 weakly excites
  17. L13:971 weakly suppresses
  18. L8:2 weakly excites
  19. L16:932 weakly excites
  20. L10:144 weakly excites
  21. L23:1266 weakly suppresses
  22. L10:49 weakly suppresses
  23. L14:54 weakly excites
  24. L15:709 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.

2 values between -1.13 and -1.052 values between -0.97 and -0.894 values between -0.81 and -0.734 values between -0.73 and -0.6510 values between -0.65 and -0.5728 values between -0.57 and -0.4962 values between -0.49 and -0.41119 values between -0.41 and -0.33234 values between -0.33 and -0.25396 values between -0.25 and -0.17596 values between -0.17 and -0.09865 values between -0.09 and -0.01828 values between -0.01 and 0.07272 values between 0.07 and 0.15127 values between 0.15 and 0.2388 values between 0.23 and 0.3160 values between 0.31 and 0.3946 values between 0.39 and 0.4722 values between 0.47 and 0.5530 values between 0.55 and 0.6321 values between 0.63 and 0.7114 values between 0.71 and 0.7917 values between 0.79 and 0.8711 values between 0.87 and 0.9510 values between 0.95 and 1.039 values between 1.03 and 1.113 values between 1.11 and 1.195 values between 1.19 and 1.274 values between 1.27 and 1.355 values between 1.35 and 1.433 values between 1.43 and 1.511 values between 1.51 and 1.593 values between 1.67 and 1.752 values between 1.75 and 1.832 values between 1.83 and 1.911 values between 1.91 and 1.992 values between 1.99 and 2.07offnegativepositiveoften

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. :-⏎
  2. .Unmarshal
  3. hesive
  4. reau
  5. 瑁
  6. شر
  7. although
  8. /screens
  9. according
  10. (models

Suppresses

  1. 自己
  2. ivated
  3. 了他的
  4. 了自己的
  5. ivating
  6. 自由
  7. rocessing
  8. agi
  9. ован
  10. himself