Neuron Explorer

Qwen3-0.6B

end-of-text token between documents

A neuron in layer 27 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 lights up on the hidden end-of-document marker that separates one web page from an unrelated one. Turning it up made the model's writing drift into odd, repetitive word loops, so the effect on text is hard to read.

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

    1, on the Mall in front of the National Air and Space Museum. The museum will also be hosting a scholarly conference on Civil War ballooning on June <|endoftext|>Most veteran peanut farmers in the Southeast will remember the devastating spread of tomato spotted wilt

  2. #2 peak

    be considered a planet, reigniting discussions.⏎In<|endoftext|>The Carnival season, or “Apokries” in Greek, is slowly coming

  3. #3 peak

    applications include improved communications technology, television displays and night-vision devices.⏎Tags: Lene Vestergaard Hau, physicist, spped of light.<|endoftext|>The Real

  4. #4 peak

    carried out for the samples with greenschist discontinuities, and found the characteristic of shear creep failure is closely related to the development of the structure surface.<|endoftext|>Japan’s Aerospace Exploration Agency (JAXA) is charting new territory in

  5. #5 peak

    on your feet. This can cause plantar fasciitis for the first time, or trigger a new bout once you have already healed.⏎If you know that<|endoftext|>CNN

  6. #6 peak

    , u, t)$.⏎<|endoftext|>While many students view summer as a time to forget about school, it remains an

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:iasmadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadowadow

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

    ’s article by Jeremy Rosen on Rav Lichtenstein is available from http://jeremyrosen.blogspot.co.uk/2015/08/rav-lichtenstein-zl.html⏎Your comment will be posted after it is approved.⏎

  2. #2 peak

    needed.⏎1. Children are starving as a result of famine. (is)⏎2. Where there is suffering that can be relieved, it should be relieved. (

  3. #3 peak

    against the Indian students. As a result, Subhash Ch<|endoftext|>BröstuligaOld Norse Dictionary - bröstuliga⏎Meaning of Old Norse word "bröstuliga" (or brǫstul

  4. #4 peak

    asti||ist, is (greek esti)|⏎|bandha (bind)||bind, bandage|⏎|iccha, isha (to wish)||wish|⏎|jan (produce offspring, born)||generate, genocide|⏎|k

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. L25:239 strongly excites
  2. L20:563 excites
  3. L19:1174 excites
  4. L21:29 excites
  5. L19:509 excites
  6. L19:1207 suppresses
  7. L14:140 suppresses
  8. L22:65 suppresses
  9. L24:77 excites opening parenthesis after a name
  10. L18:76 suppresses blank spot before a number
  11. L19:124 suppresses
  12. L23:262 suppresses
  13. L25:222 suppresses
  14. L20:433 suppresses
  15. L23:80 excites blank slot before a year or number
  16. L19:1377 excites
  17. L20:1585 excites
  18. L21:321 excites
  19. L19:963 excites
  20. L22:2621 suppresses
  21. L22:416 excites commas in dates and number lists
  22. L23:10 excites
  23. L16:1421 suppresses
  24. L16:723 suppresses

Feeds later layers

None: this is the last layer.

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 -51.34 and -49.161 values between -36.09 and -33.911 values between -29.55 and -27.371 values between -25.19 and -23.012 values between -23.01 and -20.834 values between -18.65 and -16.472 values between -16.47 and -14.292 values between -14.29 and -12.119 values between -12.11 and -9.937 values between -9.93 and -7.7512 values between -7.75 and -5.5714 values between -5.57 and -3.3935 values between -3.39 and -1.213639 values between -1.21 and 0.97104 values between 0.97 and 3.1526 values between 3.15 and 5.3312 values between 5.33 and 7.518 values between 7.51 and 9.697 values between 9.69 and 11.873 values between 11.87 and 14.054 values between 14.05 and 16.233 values between 16.23 and 18.413 values between 18.41 and 20.591 values between 20.59 and 22.771 values between 22.77 and 24.951 values between 27.13 and 29.311 values between 29.31 and 31.482 values between 31.48 and 33.662 values between 33.66 and 35.84offnegativepositiveoften

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. 三位一体
  2. 睁开
  3. 加盟
  4. 社团
  5. đạo
  6. decid
  7. tấn
  8. 木质
  9. 乙方
  10. 난

Suppresses

  1. ertain
  2. ascal
  3. amentals
  4. erval
  5. ousand
  6. irteen
  7. istency
  8. herited
  9. utions
  10. iversal