Neuron Explorer

Qwen3-0.6B

dictionary definition wording

A neuron in layer 9 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the small connecting words inside dictionary-style definitions, like 'of', 'who', 'but' or 'in' as they start to explain a meaning. It dips strongly negative on the entry markers of dictionary headings, such as the colon or 'n.' right after a word is defined. Turning it up or down did not produce a clear theme in the model's replies.

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

    usage will continue to grow as the City converts and acquires additional<|endoftext|>Yoga (Sanskrit for “tea”, or “meditation”) is a collection of psychological, spiritual, and physical exercises that originated in India. It was created to

  2. #2 peak

    the record as the highest mountain in Indonesia, the highest on the island of New Guinea and the highest on the continent of<|endoftext|>A team is a group of people who work together toward a common purpose or goal. According to the online Encyclopedia of Business

  3. #3 peak

    , labeled in Spanish.⏎Lotto Game (no words)⏎A game played like Bingo, but with pictures, not numbers. You can print out the 64 colorful game

  4. #4 peak

    the candy fifty-fifty. The robbers split the money fifty-fifty.⏎divide something (off)(from something or animals)⏎1. to separate something from something else. Let's divide the chickens off from the ducks and put the chickens in

  5. #5 peak

    TIMATE. undercarriage-See LANDING GEAR. undock-To remove an airship from its dock. useful load-See LOAD, USEFUL. valve-To release air or supporting gas from

  6. #6 peak

    century, centipede (latin centum)|⏎|asti||ist, is (greek esti)|⏎|bandha (bind)||bind, bandage|⏎|iccha, isha (to wish)||wish|⏎|

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: a person who is not very well, or a person who is not very well, or a person who is not very well. The same as a person who is not well, or a person who is not well, or a person

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

    that they engage each other during a low-speed crash to absorb crash energy.⏎What bumper means?⏎Definition of bumper (Entry 3 of 3) 1 : a device for absorbing shock or preventing damage (as in collision) specifically : a

  2. #2 peak

    !⏎3 Definitions of Rifts⏎The definition of rifts, the meaning of the word Rifts:⏎Is rifts a scrabble word? Yes!⏎- n. - A gap between cloud masses⏎- n. - A narrow fissure

  3. #3 peak

    Definition of endoplasm⏎n. - The protoplasm in the interior of a cell. 2⏎The

  4. #4 peak

    measles, and so had never developed any immunityImmunity⏎Resistance to disease, developed naturally or through exposure.. Major epidemicsEpidemic⏎Affecting or tending to affect a disproportionately large number of individuals within a population, community,

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. L7:1011 excites
  2. L5:668 suppresses
  3. L3:1896 excites
  4. L6:1838 suppresses
  5. L6:1998 excites
  6. L6:18 excites
  7. L7:1995 excites
  8. L7:122 excites
  9. L8:2351 suppresses
  10. L8:13 suppresses
  11. L8:5 suppresses
  12. L7:1150 weakly excites
  13. L3:456 weakly suppresses
  14. L8:7 weakly excites
  15. L6:13 weakly excites
  16. L7:60 weakly suppresses
  17. L7:102 weakly suppresses
  18. L7:2260 weakly suppresses
  19. L6:7 weakly suppresses
  20. L8:2354 weakly excites
  21. L5:216 weakly suppresses
  22. L7:2029 weakly suppresses
  23. L4:2541 weakly excites
  24. L6:5 weakly excites

Feeds later layers

  1. L17:2498 suppresses
  2. L11:2295 suppresses
  3. L14:1886 suppresses
  4. L13:2252 suppresses
  5. L16:1349 suppresses
  6. L14:582 suppresses
  7. L17:1443 weakly suppresses
  8. L12:2251 weakly suppresses
  9. L24:3028 weakly excites
  10. L10:2696 weakly excites
  11. L12:1877 weakly excites
  12. L13:559 weakly suppresses
  13. L11:2555 weakly suppresses
  14. L23:1625 weakly excites
  15. L10:2290 weakly excites
  16. L12:2334 weakly excites
  17. L11:937 weakly excites
  18. L16:537 weakly suppresses
  19. L14:2601 weakly suppresses
  20. L15:2309 weakly suppresses
  21. L12:2256 weakly suppresses
  22. L12:2825 weakly suppresses
  23. L25:2971 weakly excites
  24. L12:1223 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.

1 values between -4.24 and -4.111 values between -3.21 and -3.081 values between -2.70 and -2.571 values between -2.31 and -2.181 values between -2.18 and -2.062 values between -2.06 and -1.931 values between -1.93 and -1.801 values between -1.67 and -1.541 values between -1.54 and -1.413 values between -1.41 and -1.293 values between -1.16 and -1.031 values between -0.90 and -0.772 values between -0.77 and -0.647 values between -0.64 and -0.5110 values between -0.51 and -0.3955 values between -0.39 and -0.26223 values between -0.26 and -0.13938 values between -0.13 and -0.001388 values between -0.00 and 0.13778 values between 0.13 and 0.26341 values between 0.26 and 0.39108 values between 0.39 and 0.5133 values between 0.51 and 0.644 values between 0.64 and 0.774 values between 0.77 and 0.90offnegativepositiveoften

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. igned
  2. 槃
  3. jom
  4. projectId
  5. pdata
  6. ({});⏎
  7. .plist
  8. 生命周期
  9. uchs
  10. PCODE

Suppresses

  1. odef
  2. ignite
  3. defensive
  4. 颍
  5. agg
  6. ány
  7. tek
  8. öz
  9. iece
  10. 拗