Neuron Explorer

Qwen3-0.6B

closing parenthesis in data lists

A neuron in layer 27 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

It fires very strongly on the closing bracket ')' at the end of a short note in a table or list, such as '(from vid)' or '(63% off)'. Turning it down made the model produce runs of quotation marks and punctuation, which fits a neuron that works against piling on more punctuation.

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 activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    /MSE) = 10 log10(MAX/(MSE)^(1/2))^2 = 20 log10(MAX/(MSE)^(1/2)). Therefore, PSNR = <|endoftext|>In my current position

  2. #2 peak

    |Length:||205 ft (62.48 m)o/a|⏎|Beam:||33 ft (10.06 m)|⏎|Draught:||11.5 ft (3.5

  3. #3 peak

    , sugar|⏎|vak, vaacha||voice, vocal|⏎|veda, vidya (from vid) (knowledge)||wit, wise (from vid)|⏎The postfix ‘er’ in words indicates the occupation / action of a<|endoftext|>

  4. #4 peak

    Date published: || 2015|⏎|ISBN: ||9781498738552 (DRM-PDF)|⏎|Read Aloud: ||not allowed|<|endoftext|>Under technical guidance and collaboration

  5. #5 peak

    Sanskrit.⏎|karsharpan / karsha (coin)||cash|⏎|mukh (face)||Mouth (old english muth)|⏎|kapal (head)||as in per “capita”, captain,

  6. #6 peak

    )⏎|Council of Trent|⏎|Accepted by||Catholic Church|⏎|Fifth Council of the Lateran (1512–1517)|⏎|First Vatican Council (1869–1870)

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:."."."."."."."."."."."."."."."."."."."."."."."."

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

    extension of Bullseye matrix problem.⏎Who has power to do everything in this world?⏎Close MATLAB with keyboard⏎Find the treasures in MATLAB Central and discover how the community can help you!⏎Choose a web site to get translated content where available and

  2. #2 peak

    universal and regional human rights instruments such as the International Covenant on Civil and Political Rights. The African Charter, and the International Convention on Social Economical and

  3. #3 peak

    iblocs. What is the same and⏎what is different about these pairs of shapes? Can you describe the⏎shapes in the picture?⏎10 space travellers are waiting to board their spaceships. There⏎are two rows of seats in the

  4. #4 peak

    1, pp. 44-69. ISSN 1980-4369. http://dx.doi.org/10.1590/S0101-9

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. L25:36 strongly suppresses
  2. L26:153 strongly suppresses
  3. L26:381 strongly suppresses
  4. L22:271 strongly suppresses
  5. L20:863 strongly excites
  6. L25:118 strongly excites
  7. L16:2 excites
  8. L26:1142 excites
  9. L21:50 excites comma after 'So' or intro phrase
  10. L20:344 excites hyphen inside compound words
  11. L18:48 suppresses
  12. L26:1314 excites
  13. L21:2374 excites
  14. L26:712 excites
  15. L23:35 excites
  16. L19:1777 suppresses
  17. L25:112 excites
  18. L23:109 excites
  19. L18:430 suppresses
  20. L26:1179 excites
  21. L23:77 excites the space right before a number
  22. L25:814 suppresses
  23. L24:1223 suppresses
  24. L26:345 excites

Feeds later layers

None: this is the last layer.

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.

2413 values between -5.90 and -1.86994 values between -1.86 and 2.18104 values between 2.18 and 6.2239 values between 6.22 and 10.2619 values between 10.26 and 14.3018 values between 14.30 and 18.3335 values between 18.33 and 22.3738 values between 22.37 and 26.4155 values between 26.41 and 30.4544 values between 30.45 and 34.4945 values between 34.49 and 38.5223 values between 38.52 and 42.5623 values between 42.56 and 46.609 values between 46.60 and 50.6410 values between 50.64 and 54.687 values between 54.68 and 58.716 values between 58.71 and 62.754 values between 62.75 and 66.791 values between 66.79 and 70.833 values between 70.83 and 74.875 values between 74.87 and 78.903 values between 78.90 and 82.941 values between 86.98 and 91.022 values between 91.02 and 95.062 values between 95.06 and 99.091 values between 103.13 and 107.172 values between 111.21 and 115.251 values between 131.40 and 135.441 values between 151.59 and 155.63offnegativepositiveoften

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. ibern
  2. fäll
  3. ickle
  4. !”
  5. erne
  6. าม
  7. i
  8. посколь
  9. irl
  10. ,

Suppresses

  1. .
  2. "
  3. )
  4. -
  5. (
  6. *
  7. :
  8. /
  9. [
  10. ,