Neuron Explorer

Qwen3-0.6B

letters inside spelled-out words

A neuron in layer 20 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 a single letter or syllable in the middle of a word that has been split up or spelled out, such as the letters in 'C F O S U' or 'P-R-A-C-T-I-C-E', and on odd spots in non-English scripts. Turning it up did not produce a clear theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    知多少⏎Pin Yin Version Chun mian bu jue xiao Chu chu wen ti niao Ye lai feng yu sheng Hua luo zhi duo shao⏎Inspiration Lifetime Wrote about his lackadaisical

  2. #2 peak

    - To bring to a focus; to focalize; as, to focus a camera. 2⏎The word "focus" uses 5 letters: C F O S U.⏎No direct anagrams for focus found in this word list.⏎

  3. #3 peak

    Biologics in the United States Chapter 16RISK MAnAgEMEnT FOR DRugS AnD bIOLOgICS In THE unITED STATES Arcoxia, Vioxx, Acomplia, Z

  4. #4 peak

    The basic⏎source of all minerals is the hot magma in the⏎interior of the earth.⏎(more content follows the advertisement below) A D V E R T I S E M E N T⏎When magma cools,⏎

  5. #5 peak

    most recognizably in the Hebrew name of The Jewish Agency – הַסּוֹכְנוּת הַיְּהוּדִית, or simply הס

  6. #6 peak

    those questions.)⏎What’s that? You don’t understand what we mean by selecting answer A? It’s OK. You’re going to P-R-A-C-T-I-C-E and get very comfortable with the test format so you don’t have to

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: I have a 3420, and I have a 3666. The 3666 is 2136, and the 3420 is 2096. The

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

    Subhash Ch<|endoftext|>BröstuligaOld Norse Dictionary - bröstuliga⏎Meaning of Old Norse word "bröstuliga" (or brǫstul

  2. #2 peak

    Marie Barnave⏎Barnave, Antoine Pierre Joseph Marie (äNtwänˈ pyĕr zhōzĕfˈ märēˈ bärnävˈ) [key], 1761–93

  3. #3 peak

    ogd. γrīw (in Manichean and Nestorian script γryw, in Soghdian script γryw(h), γrʾyw(h) “self, body, soul” (Gharib,

  4. #4 peak

    ְדָה (vÿ’ikkavÿda), the Niphal cohortative; coming after the

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. L19:2464 excites
  2. L18:1012 suppresses
  3. L19:2076 excites
  4. L19:2081 excites
  5. L18:2405 excites
  6. L18:1457 excites
  7. L19:1133 suppresses last word before end of document
  8. L14:1424 excites
  9. L19:1502 suppresses
  10. L18:632 suppresses
  11. L19:1096 suppresses
  12. L16:1262 weakly suppresses
  13. L19:1663 weakly excites
  14. L13:148 weakly suppresses
  15. L16:104 weakly suppresses
  16. L15:2839 weakly suppresses
  17. L14:2706 weakly excites
  18. L2:105 weakly excites
  19. L14:1575 weakly excites
  20. L17:2257 weakly suppresses
  21. L12:1110 weakly suppresses
  22. L18:86 weakly suppresses
  23. L19:1491 weakly excites
  24. L14:1914 weakly suppresses

Feeds later layers

  1. L21:12 suppresses
  2. L24:367 suppresses
  3. L25:5 suppresses
  4. L23:3 suppresses
  5. L22:29 suppresses
  6. L24:7 suppresses
  7. L26:547 suppresses
  8. L27:386 suppresses
  9. L27:222 suppresses
  10. L21:142 suppresses
  11. L24:4 suppresses
  12. L27:0 weakly suppresses joins inside names and decimal digits
  13. L25:409 weakly excites
  14. L27:39 weakly suppresses
  15. L21:1149 weakly excites
  16. L27:112 weakly suppresses
  17. L27:135 weakly suppresses
  18. L27:78 weakly suppresses
  19. L22:7 weakly suppresses
  20. L24:85 weakly suppresses
  21. L25:11 weakly suppresses
  22. L27:81 weakly suppresses
  23. L26:116 weakly suppresses
  24. L22:370 weakly suppresses digits inside years like 1976

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 -7.48 and -7.012 values between -7.01 and -6.551 values between -5.63 and -5.161 values between -4.70 and -4.241 values between -4.24 and -3.782 values between -3.78 and -3.312 values between -3.31 and -2.851 values between -2.85 and -2.392 values between -2.39 and -1.933 values between -1.93 and -1.4616 values between -1.46 and -1.00178 values between -1.00 and -0.542047 values between -0.54 and -0.081519 values between -0.08 and 0.39117 values between 0.39 and 0.858 values between 0.85 and 1.312 values between 1.31 and 1.771 values between 1.77 and 2.241 values between 2.24 and 2.701 values between 3.16 and 3.621 values between 6.86 and 7.321 values between 10.56 and 11.02offnegativepositiveoften

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. :result
  2. ':[
  3. ектив
  4. '){ ⏎
  5. ){ ⏎
  6. ーム
  7. -results
  8. '; ⏎
  9. ')⏎⏎
  10. réponse

Suppresses

  1. yan
  2. ire
  3. lang
  4. iane
  5. na
  6. ern
  7. long
  8. ra
  9. irs
  10. h