Neuron Explorer

Qwen3-0.6B

end-of-document separator token

A neuron in layer 18 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 lights up on the invisible marker the model uses to show where one web document ends and another begins. When it was turned up, the written text drifted into unrelated rambling, with no clear 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

    0240.<|endoftext|>By 2003, plenty of data suggested that pesticide drift was harming

  2. #2 peak

    , 2012<|endoftext|>Some weather conditions are simply not fun to ride in. As cyclists, windy days

  3. #3 peak

    old<|endoftext|>Immune cells in the brain have surprising influence on sexual behavior⏎Relatively uncommon

  4. #4 peak

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

  5. #5 peak

    :<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'Aim High' printed from

  6. #6 peak

    :<|endoftext|>The hamstring muscles – the biceps femoris, semitendinosus,

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: about the time of the day of the way to the back of the square from to the main square, and to go there there to do to there to to there to there there to there to there to there to there to there there to

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

    asiveness of viral campaigns in social network sites: Persuasive intent as the underlying mechanism. Journal of Marketing, 18(1), 39-53. http://dx.doi.org/10.108

  2. #2 peak

    rin Froese (2008). Organic Virtue: Reading Mencius with Rousseau. Asian Philosophy 18 (1):83 – 104.⏎Tao

  3. #3 peak

    environments.⏎Source: Health Behavior and Policy Review, Volume 8, Number 4, July 2021, pp. 281-293(13)⏎Publisher: Paris Scholar Publishing Ltd.<|endoftext|>March Is

  4. #4 peak

    ⏎James Behuniak Jr (2011). Naturalizing Mencius. Philosophy East and West 61 (3):492-515.⏎David E. Soles (1999). The Nature

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. L9:81 suppresses
  2. L8:1365 suppresses
  3. L17:592 excites
  4. L4:1371 excites end-of-text marker at document breaks
  5. L2:2188 weakly excites
  6. L1:1464 weakly excites
  7. L13:1614 weakly excites 'the' after a preposition or verb
  8. L17:357 weakly suppresses
  9. L7:1106 weakly excites
  10. L5:110 weakly suppresses
  11. L0:1162 weakly suppresses
  12. L14:1709 weakly suppresses
  13. L3:2015 weakly excites
  14. L15:867 weakly suppresses
  15. L3:469 weakly excites
  16. L7:76 weakly excites
  17. L1:2966 weakly excites
  18. L1:1606 weakly excites
  19. L11:1379 weakly excites
  20. L3:1121 weakly suppresses
  21. L3:103 weakly excites
  22. L3:962 weakly suppresses
  23. L17:738 weakly excites
  24. L10:1578 weakly excites

Feeds later layers

  1. L19:2973 suppresses
  2. L21:1710 suppresses
  3. L22:2929 weakly excites
  4. L19:1272 weakly excites
  5. L20:305 weakly excites
  6. L22:546 weakly excites
  7. L23:314 weakly excites
  8. L22:2411 weakly suppresses
  9. L21:2656 weakly suppresses
  10. L21:2558 weakly suppresses
  11. L23:2747 weakly excites
  12. L20:2442 weakly suppresses
  13. L21:2077 weakly suppresses
  14. L24:1197 weakly suppresses
  15. L22:2389 weakly suppresses
  16. L22:2077 weakly suppresses
  17. L24:1028 weakly suppresses
  18. L20:532 weakly suppresses
  19. L21:2408 weakly suppresses
  20. L22:1418 weakly suppresses
  21. L19:2708 weakly suppresses mid-sentence words in web boilerplate
  22. L19:2096 weakly suppresses
  23. L25:1051 weakly suppresses
  24. L20:2760 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.

3 values between -2.57 and -2.131 values between -2.13 and -1.692 values between -1.69 and -1.2521 values between -1.25 and -0.80172 values between -0.80 and -0.363439 values between -0.36 and 0.08266 values between 0.08 and 0.522 values between 0.52 and 0.961 values between 3.17 and 3.621 values between 14.66 and 15.10offnegativepositiveoften

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. emode
  2. ocale
  3. filme
  4. cone
  5. Ụ
  6. spécialisé
  7. ត
  8. dps
  9. 번
  10. мор

Suppresses

  1. annes
  2. acular
  3. igin
  4. ект
  5. unn
  6. 纳斯
  7. ennis
  8. 那里
  9. idor
  10. nist