Neuron Explorer

Qwen3-0.6B

the 'res' fragment inside words

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 the syllable 'res' when it sits inside a longer word, such as resold, resented, resided, respiration or resurfacing. Pushed down, the model produced a stream of commas, which fits its tendency to suppress commas.

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

    them is more profitable?⏎Previously it was all simple. Every person who sold or resold something was called a trader. But nowadays everything has changed: dealers, distributors

  2. #2 peak

    wet. Plants are intolerant of root disturbance and should be planted into their permanent positions whilst still small. Division in spring is very difficult, since the plant strongly resents root disturbance. It is probably best to take quite small divisions, or basal

  3. #3 peak

    development.⏎During the early years of St. Anthony and after the merge, Fifth Street Southeast remained one of the finer streets of residence. Many of the people who resided in this neighborhood were merchant families originally from New England. The flour and milling

  4. #4 peak

    the incomparable Bill Murray as Phil Conners, an insufferable Pittsburgh weatherman. A minor local celebrity who believes himself destined for much better things, he resents

  5. #5 peak

    's allowable specifications.⏎A brake rotor has a specified minimum thickness. The brake rotor must measure thicker than the manufacturer's minimum or discard thickness after the technician has finished resurfacing the rotors. Normal braking action can cause wearing of the rotor surfaces

  6. #6 peak

    sensitive index of respiratory function, showed that mtDNA from chimpanzee, pigmy chimpanzee, and gorilla replaced the human mtDNA and restored respiration to essentially normal levels. Mitochondrial protein synthesis was also unaltered

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

    by John Kennedy at Rice University in Houston in September of 1962:⏎“We choose to go to the Moon in this decade and do the other things, not because they are easy, but because they are hard, because

  2. #2 peak

    among savages than among civilised men. Notwithstanding the differences which necessarily exist between savage and civilised ethics, the two systems must evidently have much in common, for from the days of Columbus to the present, travellers have given testimony of customs

  3. #3 peak

    object is fulfilled by the supramental being's appearance. But there is no such stress in reality: the idea of human progress itself is very probably an illusion, for there is no sign that man, once emerged from the animal stage, has

  4. #4 peak

    Street View is available, the image is from the best available vantage point looking, if possible, towards the location of the monument. Where it is not available, the satellite view is shown instead.⏎Latitude: 52.516

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. L21:344 strongly excites
  2. L21:1484 strongly excites comma after a sentence subject
  3. L20:1358 strongly excites
  4. L20:882 strongly excites
  5. L20:2082 strongly excites
  6. L21:393 strongly excites unclear: relative 'that' and list items
  7. L22:368 strongly excites
  8. L20:811 strongly suppresses
  9. L25:969 strongly suppresses
  10. L19:2493 strongly excites
  11. L24:6 strongly excites
  12. L18:1394 strongly excites
  13. L21:1070 strongly excites
  14. L21:530 strongly excites
  15. L20:610 strongly suppresses
  16. L20:173 strongly excites
  17. L20:1399 strongly excites
  18. L21:632 strongly suppresses
  19. L20:1744 strongly suppresses comma after a sentence's subject
  20. L21:1059 strongly suppresses
  21. L20:2730 strongly suppresses
  22. L21:809 strongly excites
  23. L20:2608 strongly excites
  24. L19:2645 strongly excites commas in the middle of a sentence

Feeds later layers

  1. L27:86 strongly excites
  2. L27:358 strongly suppresses
  3. L27:191 suppresses
  4. L27:77 excites
  5. L27:286 suppresses
  6. L27:130 excites
  7. L27:139 weakly excites
  8. L27:265 weakly suppresses
  9. L27:2921 weakly excites
  10. L27:936 weakly suppresses
  11. L27:947 weakly suppresses
  12. L27:575 weakly suppresses
  13. L27:347 weakly excites
  14. L27:753 weakly suppresses
  15. L27:1061 weakly excites
  16. L27:197 weakly suppresses
  17. L27:2066 weakly excites
  18. L27:1651 weakly excites
  19. L27:63 weakly excites closing parenthesis in data lists
  20. L27:168 weakly excites
  21. L27:1391 weakly suppresses
  22. L27:982 weakly suppresses
  23. L27:214 weakly excites
  24. L27:1498 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 -17.66 and -16.751 values between -15.84 and -14.931 values between -14.93 and -14.031 values between -12.21 and -11.302 values between -11.30 and -10.405 values between -9.49 and -8.589 values between -8.58 and -7.688 values between -7.68 and -6.7711 values between -6.77 and -5.8614 values between -5.86 and -4.9526 values between -4.95 and -4.0537 values between -4.05 and -3.1463 values between -3.14 and -2.23105 values between -2.23 and -1.32222 values between -1.32 and -0.42509 values between -0.42 and 0.49326 values between 0.49 and 1.40377 values between 1.40 and 2.31382 values between 2.31 and 3.21363 values between 3.21 and 4.12336 values between 4.12 and 5.03319 values between 5.03 and 5.94238 values between 5.94 and 6.84169 values between 6.84 and 7.75137 values between 7.75 and 8.6689 values between 8.66 and 9.5761 values between 9.57 and 10.4737 values between 10.47 and 11.3817 values between 11.38 and 12.2912 values between 12.29 and 13.208 values between 13.20 and 14.108 values between 14.10 and 15.019 values between 15.01 and 15.924 values between 15.92 and 16.831 values between 17.73 and 18.64offnegativepositiveoften

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. 돌
  2. sons
  3. iped
  4. artin
  5. ös
  6. сты
  7. ;b
  8. cks
  9. ilon
  10. ;n

Suppresses

  1. ,
  2. ,),
  3. ,",
  4. ,',
  5. ,’
  6. ,"
  7. ,),⏎
  8. **,
  9. ,**
  10. ,"⏎