Neuron Explorer

Qwen3-0.6B

end-of-document token and code keywords

A neuron in layer 25 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 marker that separates one web page from the next, right where a new unrelated document starts. Turning it up made the model write code words like 'if', 'while' and 'else' over and over, so this is a mixed signal.

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

    wellandgood.com.<|endoftext|>The Technology Behind Bella Hadid’s Spray on Dress⏎During the 20

  2. #2 peak

    locate.<|endoftext|>Science Fair Project Encyclopedia⏎The World Factbook⏎The World Factbook is an

  3. #3 peak

    comment.<|endoftext|>100 Chemistry Questions and Answers 8211; Basic Chemistrynbsp

  4. #4 peak

    ycetica<|endoftext|>Adding Fractions Level 3⏎The 5-step planIn Adding Fractions

  5. #5 peak

    Project:<|endoftext|>After years of experience and "combing through" catalogs⏎to find the best

  6. #6 peak

    _to_eat_gmos_so_why_are_we/<|endoftext|>A PhD can open a lot of job opportunities. Most of those jobs are academic

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 am still a student, I am still a student, I am still a student. I am still a student. I am still a student. I am still a student. I am still a student. I am still a student.

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

    print "This might take a while..."⏎> 4 for i = 1 to n⏎> 5 for j = 1 to i⏎> 6 print

  2. #2 peak

    and air conditioning systems, it also means that pollutants that originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落

  3. #3 peak

    _arr = [1, 2, 3, 4, 5] i = len(given_arr) - 1 while(i >= 0): print(given_arr[i]) i = i - 1⏎In this program

  4. #4 peak

    Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1 4 9 16 25.........⏎Basics: Divide integers to get integer outputs in all cases⏎Determine value using Look-up

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. L21:131 excites
  2. L19:1372 suppresses
  3. L17:2265 excites
  4. L18:284 suppresses
  5. L18:3070 excites
  6. L10:149 excites
  7. L21:1988 excites
  8. L22:633 suppresses
  9. L19:719 excites
  10. L17:95 suppresses
  11. L18:569 weakly excites
  12. L22:1096 weakly excites
  13. L20:1635 weakly excites
  14. L6:1800 weakly excites
  15. L19:415 weakly suppresses
  16. L14:67 weakly suppresses
  17. L22:2020 weakly excites
  18. L20:2632 weakly suppresses
  19. L23:2258 weakly excites
  20. L20:96 weakly suppresses
  21. L20:890 weakly suppresses
  22. L15:524 weakly excites
  23. L6:7 weakly suppresses
  24. L10:1995 weakly suppresses formal nouns after verbs of answering or committing

Feeds later layers

  1. L27:110 suppresses
  2. L27:1150 excites
  3. L27:791 excites
  4. L27:2076 excites
  5. L27:338 weakly excites
  6. L27:511 weakly excites
  7. L26:2856 weakly suppresses
  8. L27:94 weakly suppresses
  9. L27:201 weakly excites
  10. L27:274 weakly excites document boundary, code keywords next
  11. L27:760 weakly excites
  12. L26:82 weakly suppresses
  13. L27:463 weakly excites
  14. L27:146 weakly suppresses
  15. L27:1908 weakly excites the 'd' that starts 'die' or 'dead'
  16. L27:2484 weakly excites
  17. L26:1683 weakly suppresses
  18. L27:1666 weakly excites
  19. L27:1286 weakly excites
  20. L26:118 weakly suppresses
  21. L26:2540 weakly excites
  22. L27:1335 weakly excites
  23. L26:884 weakly suppresses
  24. L26:2462 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.

5 values between -1.42 and -1.2012 values between -1.20 and -0.9962 values between -0.99 and -0.77124 values between -0.77 and -0.55303 values between -0.55 and -0.34842 values between -0.34 and -0.122491 values between -0.12 and 0.1051 values between 0.10 and 0.314 values between 0.31 and 0.532 values between 0.53 and 0.751 values between 0.75 and 0.961 values between 0.96 and 1.181 values between 1.18 and 1.401 values between 1.40 and 1.612 values between 3.35 and 3.561 values between 3.56 and 3.781 values between 4.43 and 4.651 values between 4.86 and 5.081 values between 5.30 and 5.511 values between 5.73 and 5.951 values between 7.03 and 7.25offnegativepositiveoften

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. else
  2. else
  3. try
  4. try
  5. while
  6. raise
  7. assert
  8. while
  9. break
  10. return

Suppresses

  1. ankind
  2. imated
  3. inent
  4. rending
  5. ừa
  6. lemen
  7. ennessee
  8. étant
  9. /browse
  10. lotte