Neuron Explorer

Qwen3-0.6B

ellipsis cut-off in quotes

A neuron in layer 26 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

This neuron fires on an ellipsis (three dots) placed where a quotation or excerpt has been cut short, like "wheel of his cart[...]". Turned down, the model repeated ".\"" and dots; turned up, it produced "</" tags and odd words.

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

    nature to desire truth, goodness and beauty, and those are relevant to all humans in every culture, in every country, in every age.⏎READ ON BELOW…Don’t Sneer at Truth, Beauty and Goodness — You Were Made for

  2. #2 peak

    physical<|endoftext|>Energy Transition has a look at the German "Energy Transition" to renewable energy and some of the misconceptions spread about it by its critics - Angst… that the Energiewende will work.⏎Unlike the Swedes and Italians, the

  3. #3 peak

    The Preamble was several full clauses in the Articles of Confederation. Perhaps using the abbreviated version in the Preamble was saving too much paper. Just sayin….⏎We the People of the United States, in Order to form a more perfect

  4. #4 peak

    , and its height was three cubits. …Now each stand had four bronze wheels with bronze axles… The four wheels were underneath the borders, and the axles of the wheels were

  5. #5 peak

    odus 14:25 NASB⏎…the wheel of his cart… —Isaiah 28:28 NASB excerpt⏎A horizontal

  6. #6 peak

    moving to a new web site. There may be some disruption of service from 3/23/13-3/24/13<|endoftext|>…In 197

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

    ke Ferdinand, and to the Emperor Maximilian II.⏎Mattioli's ` Commentarii in sex libros Pedacii Dioscoridis,' his chef-d'oeuvre, the gradual production and improvement of which occupied his leisure hours throughout his

  2. #2 peak

    prevented!** The current recommendations from the 2005 Dietary Guidelines for Americans suggests that general, healthy members of the population consume no more than 2,300 mg of sodium.*** The American Heart Association recommends no more than

  3. #3 peak

    time and distinguishing characteristics of the human psyche linked to the heavens above.⏎In the mid-eighteenth century the Qianlong Emperor, ruler of the Qing (Ch’ing) dynasty—a Manchu (not Chinese) by birth—was fascinated by

  4. #4 peak

    their adult lives.⏎Over 16,250 genes were found within the 516Mb genome (the human genome has about 23,000 genes over 3Gb of DNA). Genes were found corresponding

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. L21:1894 strongly excites
  2. L25:118 strongly excites
  3. L2:202 strongly suppresses "of" after "many" or "which"
  4. L19:89 strongly suppresses
  5. L18:48 strongly suppresses
  6. L16:41 strongly excites
  7. L0:117 excites
  8. L20:863 excites
  9. L21:173 excites
  10. L21:571 excites
  11. L23:35 excites
  12. L2:1753 excites
  13. L15:123 suppresses
  14. L20:8 suppresses
  15. L14:51 excites colon that introduces a statement
  16. L5:621 suppresses
  17. L9:7 excites
  18. L17:38 suppresses
  19. L20:2984 suppresses
  20. L16:170 suppresses
  21. L16:39 excites
  22. L16:86 excites
  23. L13:10 excites
  24. L0:1127 suppresses

Feeds later layers

  1. L27:1053 strongly suppresses
  2. L27:1801 suppresses
  3. L27:692 excites 'Ch' starting a proper name
  4. L27:487 excites
  5. L27:433 suppresses
  6. L27:136 suppresses
  7. L27:986 suppresses
  8. L27:66 excites
  9. L27:775 suppresses
  10. L27:114 excites
  11. L27:445 suppresses
  12. L27:1396 suppresses
  13. L27:841 weakly suppresses
  14. L27:2710 weakly excites
  15. L27:2098 weakly suppresses
  16. L27:2439 weakly excites slash inside web addresses
  17. L27:1706 weakly suppresses
  18. L27:637 weakly suppresses
  19. L27:2957 weakly excites
  20. L27:2777 weakly suppresses
  21. L27:1418 weakly excites
  22. L27:1078 weakly excites
  23. L27:2946 weakly excites
  24. L27:598 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 -16.61 and -15.181 values between -15.18 and -13.751 values between -12.33 and -10.901 values between -10.90 and -9.471 values between -9.47 and -8.041 values between -8.04 and -6.622 values between -6.62 and -5.192 values between -5.19 and -3.761 values between -3.76 and -2.33122 values between -2.33 and -0.903377 values between -0.90 and 0.5286 values between 0.52 and 1.9553 values between 1.95 and 3.3829 values between 3.38 and 4.8121 values between 4.81 and 6.2313 values between 6.23 and 7.6613 values between 7.66 and 9.0918 values between 9.09 and 10.5211 values between 10.52 and 11.9514 values between 11.95 and 13.379 values between 13.37 and 14.8022 values between 14.80 and 16.2317 values between 16.23 and 17.6614 values between 17.66 and 19.0814 values between 19.08 and 20.5110 values between 20.51 and 21.9410 values between 21.94 and 23.3714 values between 23.37 and 24.807 values between 24.80 and 26.223 values between 26.22 and 27.655 values between 27.65 and 29.081 values between 29.08 and 30.517 values between 30.51 and 31.931 values between 31.93 and 33.362 values between 33.36 and 34.791 values between 34.79 and 36.221 values between 36.22 and 37.652 values between 39.07 and 40.50offnegativepositiveoften

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. ."⏎⏎⏎
  2. ."</
  3. .</
  4. ?"⏎⏎⏎⏎
  5. ."},⏎
  6. .'</
  7. ).⏎⏎⏎
  8. ?</
  9. '</
  10. ."⏎⏎⏎⏎

Suppresses

  1. ⏎⏎
  2. ⏎ ⏎
  3. ⏎
  4. ⏎ ⏎
  5. ⏎ ⏎
  6. ⏎ ⏎
  7. ⏎ ⏎
  8. ⏎→⏎
  9. "⏎⏎
  10. ⏎ ⏎