Neuron Explorer

Qwen3-0.6B

end of document token

A neuron in layer 12 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 mostly switches on at the invisible marker that separates one web page from the next, where a new, unrelated text begins. Turning it up made the model's writing wander and repeat itself, without a clear theme.

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

    a few steps, I would suggest you go through this: https://www.thepythoncode.com/article/hide-secret-data-in-images-using-steganography-python<|endoftext|>Acute paranoia is a clinically recognized personality disorder characterized by extreme distrust, hostility and

  2. #2 peak

    made of four 1N4007 PN junction diodes and a filter capacitor of 2.2uF/450V.⏎This<|endoftext|>Warning: Use of undefined constant user_level - assumed 'user_level' (this

  3. #3 peak

    data from TC1.<|endoftext|>The concept of welfare state⏎We all know that most of the modern democratic countries

  4. #4 peak

    firstname.lastname@example.org.<|endoftext|>The Musical Representation⏎How human musical experience emerges from the audition of organized tones is

  5. #5 peak

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

  6. #6 peak

    Statistics software.⏎The software version addressed is SAS LASR Analytics Server 2.5.⏎SAS In-Memory Statistics Overview<|endoftext|>Thomas Edison Pitch Contest⏎The Thomas Edison Pitch Contest was created in 20

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: the need to know, how to apply and then when to use it correctly). Answer, Question: What is the key human resource management information system approach for this question where to adapt the best one? ? Answering process: You need to

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

    a matrix⏎Find nth maximum⏎Matlab Basics II - Create a vector with a repeated entry⏎Create sequnce 1

  2. #2 peak

    /rural/2017-05-23/the-future-of-automation-in-the-mining-industry/8550636<|endoftext|>Service Oriented Architecture⏎Service-oriented architecture (SOA) is an evolution

  3. #3 peak

    from China, snow peas from Guatemala and asparagus from Peru.⏎As fears of mad cow

  4. #4 peak

    is determined" (Zuck, pp. 20). F. Biblical Theology "Biblical theology is the study of divine revelation as File Size: KB.⏎Get this from a library! Principles of biblical interpretation: (sac

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. L4:121 excites
  2. L10:157 suppresses end of document and web boilerplate
  3. L7:1625 suppresses
  4. L10:1604 suppresses
  5. L9:835 excites
  6. L9:1301 excites
  7. L9:232 suppresses
  8. L7:885 weakly excites
  9. L9:260 weakly suppresses
  10. L9:1712 weakly suppresses
  11. L11:1810 weakly excites
  12. L10:1879 weakly excites
  13. L10:890 weakly excites
  14. L10:507 weakly suppresses
  15. L7:45 weakly suppresses
  16. L11:723 weakly suppresses
  17. L7:1308 weakly suppresses
  18. L2:1549 weakly suppresses
  19. L9:414 weakly suppresses
  20. L4:761 weakly suppresses
  21. L9:128 weakly suppresses
  22. L2:924 weakly excites
  23. L5:11 weakly excites
  24. L7:1146 weakly excites

Feeds later layers

  1. L13:283 suppresses
  2. L14:2556 weakly suppresses
  3. L16:275 weakly suppresses
  4. L14:905 weakly suppresses
  5. L20:2397 weakly suppresses
  6. L17:857 weakly excites
  7. L20:52 weakly excites
  8. L20:452 weakly excites
  9. L14:2030 weakly excites
  10. L22:12 weakly excites
  11. L13:2946 weakly suppresses
  12. L23:57 weakly excites
  13. L15:1686 weakly suppresses
  14. L14:84 weakly suppresses
  15. L21:45 weakly excites
  16. L18:151 weakly excites
  17. L19:25 weakly excites
  18. L15:2628 weakly suppresses
  19. L23:143 weakly excites
  20. L15:145 weakly excites
  21. L17:458 weakly excites
  22. L21:114 weakly excites
  23. L16:248 weakly excites
  24. L15:1810 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.

1 values between -2.36 and -2.271 values between -2.27 and -2.171 values between -1.60 and -1.501 values between -1.21 and -1.121 values between -1.02 and -0.931 values between -0.93 and -0.831 values between -0.83 and -0.743 values between -0.74 and -0.643 values between -0.64 and -0.544 values between -0.54 and -0.456 values between -0.45 and -0.3527 values between -0.35 and -0.26106 values between -0.26 and -0.16407 values between -0.16 and -0.071082 values between -0.07 and 0.031165 values between 0.03 and 0.13651 values between 0.13 and 0.22326 values between 0.22 and 0.3286 values between 0.32 and 0.4121 values between 0.41 and 0.517 values between 0.51 and 0.602 values between 0.60 and 0.701 values between 0.79 and 0.891 values between 0.89 and 0.993 values between 1.37 and 1.46offnegativepositiveoften

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. trace
  2. fee
  3. atory
  4. other
  5. Trace
  6. matter
  7. enu
  8. :⏎⏎
  9. ırken

Suppresses

  1. sexist
  2. tits
  3. flank
  4. UserService
  5. ysize
  6. ingly
  7. Nonnull
  8. ')</
  9. ']}'
  10. itives