Neuron Explorer

Qwen3-0.6B

end of text after an abrupt cut

A neuron in layer 15 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 invisible end-of-document marker, mostly when the text before it stops in the middle of a thought or right after a number. Turning it up or down did not change the output in any clear way.

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

    tried to do this multiple times but I'm really confused about the equations and when to apply them.<|endoftext|>Plant Holiday Amaryllis Bulbs⏎If you had a blooming am

  2. #2 peak

    to review it. We're adding the slope at a single point to the slope over a big area, which is something I can't quite work out visually.<|endoftext|>The world produces more than enough food to feed everyone, yet each night more than

  3. #3 peak

    , 2012<|endoftext|>Some weather conditions are simply not fun to ride in. As cyclists, windy days

  4. #4 peak

    65-84 age group.⏎<|endoftext|>An unstructured genre that blends high aesthetic standards with nonfiction commentary, the journalistic

  5. #5 peak

    ,0<|endoftext|>Guidelines for Restoring Cultural Landscapes⏎Identifying, retaining and preserving

  6. #6 peak

    24.12 million.<|endoftext|>Dr. Smith, a family medicine physician, is preparing for a full day of

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: "It was a special time of the day, and the time of the day and the time of the day of what it would be, and the time the time of the day, the time of the day, and the time of the

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

    What is the highest overlook on the rim of the Grand Canyon?⏎Point Imperial, 8,

  2. #2 peak

    What is the difference between 3-axis and 5-axis?⏎Three-axis machining is processed by

  3. #3 peak

    4⏎1 T T F T⏎2 T T T F⏎3 F T T F⏎4 T F F T<|endoftext|>July 7, 2016⏎Experimenting with Post-Quantum Cryptography⏎Quantum computers

  4. #4 peak

    experiences right at home. In this article, we will look at five strategies for motivating students and keeping their brains engaged over the summer.⏎

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. L10:507 suppresses
  2. L8:808 weakly excites
  3. L6:254 weakly suppresses
  4. L6:1911 weakly suppresses
  5. L0:3047 weakly excites
  6. L12:141 weakly excites
  7. L7:2936 weakly excites
  8. L6:64 weakly suppresses
  9. L14:2348 weakly excites
  10. L10:271 weakly suppresses
  11. L6:1273 weakly suppresses open parenthesis in citations
  12. L6:492 weakly excites
  13. L11:1325 weakly suppresses
  14. L11:954 weakly suppresses
  15. L4:2843 weakly excites
  16. L6:1089 weakly suppresses
  17. L9:2578 weakly suppresses
  18. L12:1705 weakly excites
  19. L12:1600 weakly suppresses
  20. L7:302 weakly suppresses
  21. L8:148 weakly excites
  22. L0:1369 weakly suppresses
  23. L10:1095 weakly suppresses
  24. L12:178 weakly suppresses

Feeds later layers

  1. L22:2857 weakly suppresses
  2. L24:2207 weakly suppresses
  3. L27:1764 weakly suppresses
  4. L23:3039 weakly suppresses prepositions in how-to location phrases
  5. L24:2498 weakly suppresses unclear: word before an event or name
  6. L22:1552 weakly suppresses
  7. L23:2691 weakly suppresses
  8. L20:1989 weakly excites
  9. L20:1047 weakly suppresses
  10. L24:730 weakly suppresses
  11. L18:400 weakly suppresses
  12. L17:1072 weakly excites
  13. L22:2377 weakly suppresses
  14. L23:520 weakly suppresses
  15. L21:55 weakly excites
  16. L16:1709 weakly excites
  17. L23:1381 weakly suppresses
  18. L18:1137 weakly suppresses
  19. L23:2539 weakly suppresses
  20. L22:2348 weakly suppresses
  21. L21:2464 weakly suppresses
  22. L20:240 weakly excites
  23. L20:1076 weakly suppresses
  24. L20:14 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 -1.37 and -1.181 values between -0.99 and -0.791 values between -0.79 and -0.6019 values between -0.60 and -0.41105 values between -0.41 and -0.21733 values between -0.21 and -0.021708 values between -0.02 and 0.181048 values between 0.18 and 0.37261 values between 0.37 and 0.5621 values between 0.56 and 0.762 values between 0.95 and 1.141 values between 1.14 and 1.341 values between 1.34 and 1.531 values between 2.50 and 2.691 values between 3.47 and 3.661 values between 4.24 and 4.441 values between 4.44 and 4.631 values between 6.18 and 6.37offnegativepositiveoften

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. ertz
  2. migrate
  3. Frage
  4. ette
  5. ')]⏎
  6. iker
  7. esthetic
  8. udit
  9. edia
  10. дер

Suppresses

  1. 县委
  2. cxx
  3. ifornia
  4. actories
  5. 社会各界
  6. AppComponent
  7. Belgi
  8. 同志们
  9. 干警
  10. 昼