Neuron Explorer

Qwen3-0.6B

line breaks in web footers and metadata

A neuron in layer 19 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the line break at the end of a line in web page leftovers, such as link lists, citation metadata, or 'You may also like' footers. It turned negative at the end-of-text marker. Turning it up made the model write site-style filler such as 'Click to see more', which fits the theme but is not conclusive.

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

    s = session.save(k);⏎In this same case, persist() will never give any value back to the client, hope you are clear.⏎You Might Also Like⏎::. About the Author .::<|endoftext|>Jamestown Elementary hosts an elementary Montessor

  2. #2 peak

    2, from http://www.studymode.com/essays/The-Struggle-For-Power-In-The-994323.html⏎"The Struggle for Power in the Ussr" StudyMode.com.

  3. #3 peak

    shows pages 1–2. Sign up to view the full content.⏎This preview has intentionally blurred sections. Sign up to view the full version.View Full Document⏎Unformatted

  4. #4 peak

    them what inspires them and how they want to change the world. She works with the Hindu-BusinessLine newspaper, and lives in Chennai.⏎You may also like...⏎- Enviro-Challenge, Second Edition: a compilation of the best of Green

  5. #5 peak

    |Journal||Veterinarija ir Zootechnika|⏎|Publication status||Published - 17 Oct 2008|⏎- age-related changes⏎- equid skulls⏎- horse use⏎- work-related changes<|endoftext|>Make your

  6. #6 peak

    in which it appears.”⏎For more information, The report card can be found online: www.nationsreportcard.gov.MORE IN Vermont News⏎- Most Popular⏎- Most Emailed⏎- MEDIA GALLERY<|endoftext|>Chim's striking image

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: the problem is, we're in the right moment of the 2nd decade of the 2nd century, and the entire 2nd century is a key time of focus, and the focus is on the developing of the problem,

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

    , each of which is shown in turn below:⏎The distance between any grid point and the rectangular border is one unit of measurement (equal to the length of the<|endoftext|>Shree Ganeshay Namah⏎How to write Anchoring Script in English

  2. #2 peak

    Project:<|endoftext|>After years of experience and "combing through" catalogs⏎to find the best

  3. #3 peak

    two different numbers cant have the same cube). Therefore, by Definition 1, is one-to-one....⏎View Full Document⏎- Spring '09<|endoftext|>part of speech: noun⏎definition 1: an upright line or pole with

  4. #4 peak

    etto is already<|endoftext|>The North Carolina Gold Rush took place decades prior to the 1849

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. L16:2046 excites
  2. L16:235 suppresses
  3. L17:658 suppresses
  4. L17:1088 excites
  5. L18:926 excites
  6. L16:1371 excites
  7. L15:774 excites
  8. L12:2927 suppresses
  9. L17:424 suppresses sentence-ending full stops
  10. L15:1845 excites
  11. L16:2772 excites
  12. L10:1575 weakly excites
  13. L18:742 weakly excites
  14. L14:2071 weakly suppresses
  15. L18:2322 weakly suppresses end-of-text token after a finished passage
  16. L17:1783 weakly excites
  17. L1:2224 weakly excites 'of' after 'died' (cause of death)
  18. L15:2675 weakly excites
  19. L12:2368 weakly suppresses
  20. L18:412 weakly suppresses
  21. L16:1411 weakly excites
  22. L17:1502 weakly suppresses
  23. L6:749 weakly suppresses
  24. L17:2723 weakly excites

Feeds later layers

  1. L21:2511 excites
  2. L20:1381 excites
  3. L22:717 excites
  4. L22:2237 excites
  5. L20:2903 excites
  6. L21:2834 weakly excites
  7. L22:468 weakly excites
  8. L26:497 weakly excites
  9. L21:598 weakly excites
  10. L25:732 weakly excites
  11. L25:1638 weakly excites
  12. L22:263 weakly excites
  13. L22:1466 weakly excites
  14. L22:17 weakly excites
  15. L21:1537 weakly excites end of document boundary token
  16. L21:1654 weakly excites
  17. L27:2028 weakly excites
  18. L27:1936 weakly excites
  19. L23:1821 weakly excites
  20. L22:1797 weakly suppresses
  21. L27:1608 weakly excites
  22. L20:2452 weakly excites
  23. L26:162 weakly suppresses
  24. L24:329 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 -11.09 and -10.591 values between -5.10 and -4.601 values between -4.10 and -3.602 values between -3.60 and -3.101 values between -2.60 and -2.103 values between -2.10 and -1.604 values between -1.60 and -1.10109 values between -1.10 and -0.601516 values between -0.60 and -0.102090 values between -0.10 and 0.40137 values between 0.40 and 0.9023 values between 0.90 and 1.407 values between 1.40 and 1.894 values between 1.89 and 2.395 values between 2.39 and 2.891 values between 3.39 and 3.891 values between 3.89 and 4.391 values between 6.39 and 6.891 values between 8.38 and 8.88offnegativepositiveoften

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. -end
  2. -END
  3. END
  4. 下一篇
  5. END
  6. 文章
  7. Related
  8. ENDED
  9. RELATED
  10. 评论

Suppresses

  1. localVar
  2. RuleContext
  3. ổi
  4. ineTransform
  5. SWEP
  6. 摆在
  7. icular
  8. gameObject
  9. decis
  10. setQuery