Neuron Explorer

Qwen3-0.6B

document boundary 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 fires most on the invisible 'end of text' marker that separates one web page or document from the next. It also fires weakly on a few words that start a new idea. Turned up, the model's output became repetitive and unfocused.

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

    are checked to verify they meet the required specifications.⏎<|endoftext|>RASPBERRY Pi - LEARNING HOW TO MAKE YOUR ROBOT WORK⏎

  2. #2 peak

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

  3. #3 peak

    Hospitals use it to access patient records and share test results. Governments use it for elections, services and disaster alerts. The global financial system—bank transfers,

  4. #4 peak

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

  5. #5 peak

    to give updated information once medical findings become available.<|endoftext|>History of Women Cheesemakers⏎Throughout human history, women were the primary

  6. #6 peak

    on the book "Dots and Dashes: Adventures at the Dawn of the Wireless Age" is at http://vubooks.com<|endoftext|>Child health nursing is designed to help the students to develop and understand the concept 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: the more I have been thinking about, the more I have been thinking about it. So, I have to try to find something in it, and so on. In other words, the more I have been thinking about, the more I

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

    rih B. Approach to family assessment and intervention. Natl Med J India 2017;30:27

  2. #2 peak

    the Aurora Borealis. It contained standard scientific instruments.⏎But things - almost - went horribly wrong.⏎Routine

  3. #3 peak

    the discs that cushion the bones of the spine (the vertebrae). To treat degenerative disc disease, doctors usually begin with conservative (nonsurgical

  4. #4 peak

    Canada, by the British North America Act (July 1, 1867). The province of Canada is divided into Ontario and Quebec. The United States of America

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. L10:369 strongly suppresses
  2. L9:1033 suppresses unclear: end-of-text markers and word fragments
  3. L7:1053 suppresses
  4. L11:784 suppresses
  5. L11:607 excites
  6. L7:1130 excites
  7. L10:762 suppresses
  8. L4:1411 excites
  9. L8:1515 suppresses
  10. L6:927 suppresses
  11. L11:1143 suppresses
  12. L9:1131 suppresses
  13. L11:203 suppresses
  14. L4:89 suppresses
  15. L10:258 weakly excites
  16. L0:1128 weakly suppresses
  17. L7:911 weakly suppresses
  18. L10:1846 weakly excites
  19. L8:489 weakly excites
  20. L0:999 weakly suppresses
  21. L5:442 weakly suppresses
  22. L1:981 weakly excites
  23. L0:1867 weakly suppresses
  24. L7:746 weakly excites

Feeds later layers

  1. L13:1481 suppresses
  2. L15:606 suppresses
  3. L13:201 suppresses
  4. L17:320 suppresses
  5. L16:539 suppresses
  6. L18:53 suppresses
  7. L15:1814 weakly excites
  8. L14:969 weakly suppresses
  9. L15:794 weakly suppresses
  10. L15:2218 weakly excites
  11. L13:1681 weakly excites
  12. L13:1036 weakly excites
  13. L15:1034 weakly excites
  14. L16:2398 weakly excites
  15. L13:3032 weakly excites
  16. L14:1710 weakly suppresses
  17. L14:2982 weakly excites
  18. L13:1316 weakly suppresses
  19. L19:159 weakly suppresses
  20. L17:2031 weakly excites
  21. L19:393 weakly suppresses
  22. L19:55 weakly suppresses
  23. L22:21 weakly suppresses
  24. L18:44 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 -4.06 and -3.882 values between -3.88 and -3.701 values between -3.70 and -3.521 values between -3.52 and -3.332 values between -3.15 and -2.974 values between -2.97 and -2.795 values between -2.79 and -2.619 values between -2.61 and -2.435 values between -2.43 and -2.254 values between -2.25 and -2.078 values between -2.07 and -1.898 values between -1.89 and -1.7017 values between -1.70 and -1.5216 values between -1.52 and -1.3420 values between -1.34 and -1.1624 values between -1.16 and -0.9834 values between -0.98 and -0.8050 values between -0.80 and -0.6258 values between -0.62 and -0.44159 values between -0.44 and -0.25455 values between -0.25 and -0.071862 values between -0.07 and 0.11697 values between 0.11 and 0.29274 values between 0.29 and 0.4796 values between 0.47 and 0.6546 values between 0.65 and 0.8320 values between 0.83 and 1.0111 values between 1.01 and 1.209 values between 1.20 and 1.385 values between 1.38 and 1.562 values between 1.56 and 1.741 values between 1.74 and 1.921 values between 1.92 and 2.101 values between 3.01 and 3.19offnegativepositiveoften

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. genie
  2. risen
  3. resses
  4. bas
  5. odes
  6. encing
  7. 觐
  8. ulling
  9. Genesis
  10. ouver

Suppresses

  1. 脂
  2. -badge
  3. herself
  4. otive
  5. alary
  6. enda
  7. Câmara
  8. ạt
  9. .HTML
  10. otos