Neuron Explorer

Qwen3-0.6B

unclear: text just before document breaks

A neuron in layer 13 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 is weak and scattered. Its highest readings come at the points where one web page ends and the next begins, but the pattern is loose and the steering output shows no clear effect.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    ]⏎[ back to Sever Tipei's home page | Computer Music Project | Music Composition and Scientific Sonification Project ]<|endoftext|>We all face grief in life and each individual handles grief differently. Grief has

  2. #2 peak

    the behavior of websites, computer programs, various online applications, etc. Namely the execution of commands within program files makes the web as dynamic and flexible as we know it today. Files are also used for data processing purposes. Data files store data

  3. #3 peak

    , then The Gratitude Project is the book for you.⏎Reviewed by Lucy Boomer, People & Organisational Development Manager<|endoftext|>Priests and People (UK). 6 pp.⏎IT IS A FEW

  4. #4 peak

    is much cooler than explaining it.⏎Thanks for Subscribing!⏎You'll need to confirm your email, so please check your inbox.⏎(click anywhere to close)<|endoftext|>According to findings of a major HIV prevention trial presented Monday, March 4,

  5. #5 peak

    the center.⏎By blowing the air between the books, you have an increase in velocity and a decrease in pressure, thus a net downward force on the paper.<|endoftext|>On May 28, 1828, began the official life

  6. #6 peak

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

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 was told, "I was told, "I was told, "I was told, "I was told, "I was told, "I was told, "I was told, "I was told, "I was told

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

    ricular ejection fraction falls, the risk of SCD increases regardless of the underlying cause of the heart disease. Ironically, CHF patients who receive certain types of antiarrhythmic drugs are particularly

  2. #2 peak

    list of goals and check them off as you complete them. This could help you be more responsible in school.⏎- Another use of cell phones – Isn't it amazing how many features there are on a cell phone? Not just for staying in touch

  3. #3 peak

    source code for the Diebold TS and TSx voting machines. She recovered the code from the Diebold website. 1 What unfolded was a fascinating chronicle of corporate irresponsibility

  4. #4 peak

    ilisa's fourth birthday I started to analyse one particular class of Lunda-designs. As these designs turned out to have some interesting properties I gave them the name of Liki-designs. To the left are two examples.⏎How may we

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. L12:284 suppresses
  2. L11:346 excites
  3. L9:720 excites
  4. L12:2210 excites
  5. L9:255 excites
  6. L11:2097 excites
  7. L11:236 excites
  8. L12:360 suppresses
  9. L8:220 excites
  10. L10:2194 excites
  11. L9:507 excites
  12. L11:2337 excites
  13. L7:2153 excites
  14. L11:1985 weakly suppresses
  15. L8:1723 weakly suppresses
  16. L8:406 weakly excites word after a question starter
  17. L12:320 weakly suppresses
  18. L3:435 weakly suppresses
  19. L7:1183 weakly suppresses
  20. L12:1720 weakly suppresses
  21. L10:408 weakly suppresses
  22. L10:867 weakly excites
  23. L12:220 weakly excites
  24. L8:1279 weakly suppresses

Feeds later layers

  1. L15:919 excites
  2. L14:2409 excites
  3. L18:1932 excites
  4. L14:2753 suppresses
  5. L14:1933 excites
  6. L20:1396 suppresses
  7. L14:1448 weakly suppresses
  8. L16:504 weakly excites
  9. L14:1300 weakly excites
  10. L16:1308 weakly suppresses
  11. L21:1531 weakly suppresses
  12. L14:2029 weakly excites
  13. L27:1123 weakly excites
  14. L17:2818 weakly suppresses
  15. L14:2429 weakly suppresses
  16. L14:2077 weakly excites
  17. L14:3031 weakly excites
  18. L15:2037 weakly excites
  19. L14:1866 weakly excites
  20. L15:893 weakly excites
  21. L20:1967 weakly excites
  22. L15:639 weakly suppresses
  23. L18:2719 weakly suppresses
  24. L19:2604 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.

2 values between -1.55 and -1.421 values between -1.42 and -1.294 values between -1.29 and -1.163 values between -1.16 and -1.0310 values between -1.03 and -0.9014 values between -0.90 and -0.7726 values between -0.77 and -0.6424 values between -0.64 and -0.5154 values between -0.51 and -0.38140 values between -0.38 and -0.25305 values between -0.25 and -0.121560 values between -0.12 and 0.011213 values between 0.01 and 0.14324 values between 0.14 and 0.26110 values between 0.26 and 0.3956 values between 0.39 and 0.5221 values between 0.52 and 0.6513 values between 0.65 and 0.7811 values between 0.78 and 0.915 values between 0.91 and 1.044 values between 1.17 and 1.302 values between 1.30 and 1.431 values between 2.08 and 2.211 values between 2.21 and 2.341 values between 2.59 and 2.721 values between 3.11 and 3.241 values between 3.24 and 3.371 values between 3.50 and 3.63offnegativepositiveoften

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. imb
  2. stk
  3. 垧
  4. ướ
  5. 上下
  6. üt
  7. pickups
  8. }*/⏎
  9. 贸
  10. ipeg

Suppresses

  1. realised
  2. ENDOR
  3. 见证
  4. ocabulary
  5. GroupName
  6. fate
  7. 很久
  8. existence
  9. ardo
  10. doctype