Neuron Explorer

Qwen3-0.6B

past participles after passive or auxiliary verbs

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron responds to past-tense verb forms that follow a helper word, such as "was taken", "is published", or "was created". It is strongly negative on "been" in phrases like "has been designed". Turning it up or down changed the continuations into unrelated text, so the steering doesn't show 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

    .⏎Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover

  2. #2 peak

    dogs, and children.⏎Having said this, if you handle the plant properly and keep it away from kids and pets, this is one of the easiest plants to take care of, no matter how

  3. #3 peak

    the popularis politician Gaius Sempronius Gracchus in 123 B.C.; it remained an object of political contention until it was taken under the control of the autocratic Roman emperors.⏎- List of Latin

  4. #4 peak

    the melt, relying on the gas partial pressure difference and the principle of surface adsorption, absorb hydrogen in the melt, adsorb oxidized slag, and are taken out of the melt surface as the bubbles rise, so that the melt can be

  5. #5 peak

    ) saw potential in his work, and began mass-marketing them. He created several versions over the years, the last and most popular of which was this one created in 1941.⏎Christ at Heart's Door. The lightly concealed

  6. #6 peak

    ? He answered the question himself: It is Atheism!⏎Before long, especially in the South and (as the country expanded) in the West, evolution was taken to be the equivalent of godlessness. The great evangelist Dwight Moody—the Billy

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 have been, by my authoring, [s) [s) [s) [s) [s) [s) [s) [s) [s) [s) [s) [s) [s)

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

    Island.⏎The New Zealand Hydrological Society⏎This article has been published in Journal of Hydrology (NZ). © 1995 The

  2. #2 peak

    Using this item⏎This item has been provided for private study purposes (such as school projects, family and local history research

  3. #3 peak

    should always use the weight and type of motor oil as specified by your vehicle’s manufacturer as noted in your vehicle’s owner’s manual. Your vehicle’s engine has been designed to exact standards and to perform in a manner that is designed to give it

  4. #4 peak

    132.5 East (227.5 West). 38 meter/pixel resolution.⏎Note: this THEMIS visual image has not been radiometrically nor geometrically calibrated for this preliminary release. An empirical correction has

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. L8:255 suppresses
  2. L16:2227 suppresses
  3. L16:1602 excites
  4. L9:2831 excites
  5. L2:2539 excites
  6. L17:1627 excites
  7. L17:1012 excites
  8. L14:2453 suppresses
  9. L3:1153 suppresses
  10. L4:135 weakly suppresses
  11. L11:532 weakly excites verb after a gerund subject
  12. L12:2964 weakly suppresses
  13. L6:1103 weakly suppresses
  14. L6:1741 weakly suppresses
  15. L10:668 weakly suppresses
  16. L5:660 weakly suppresses
  17. L6:1621 weakly suppresses
  18. L11:403 weakly suppresses
  19. L10:388 weakly suppresses
  20. L6:2482 weakly suppresses
  21. L4:945 weakly excites
  22. L16:1532 weakly suppresses
  23. L10:1507 weakly excites
  24. L2:2369 weakly excites

Feeds later layers

  1. L21:645 excites
  2. L19:1151 suppresses
  3. L21:1047 excites
  4. L22:548 weakly suppresses
  5. L19:2895 weakly excites
  6. L19:2331 weakly excites
  7. L24:265 weakly suppresses
  8. L22:1370 weakly excites
  9. L20:1253 weakly excites
  10. L23:271 weakly suppresses
  11. L19:652 weakly suppresses
  12. L21:1514 weakly suppresses
  13. L20:650 weakly suppresses
  14. L20:850 weakly excites
  15. L23:511 weakly excites
  16. L21:2788 weakly excites
  17. L21:2218 weakly excites
  18. L27:787 weakly excites
  19. L23:592 weakly excites
  20. L24:1510 weakly excites
  21. L23:2181 weakly suppresses
  22. L24:150 weakly excites
  23. L23:469 weakly excites
  24. L23:548 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.

1 values between -28.38 and -27.451 values between -21.92 and -21.001 values between -11.77 and -10.851 values between -9.00 and -8.082 values between -8.08 and -7.161 values between -7.16 and -6.245 values between -6.24 and -5.315 values between -5.31 and -4.3914 values between -4.39 and -3.4720 values between -3.47 and -2.5516 values between -2.55 and -1.6293 values between -1.62 and -0.702776 values between -0.70 and 0.22803 values between 0.22 and 1.1483 values between 1.14 and 2.0736 values between 2.07 and 2.9920 values between 2.99 and 3.9113 values between 3.91 and 4.836 values between 4.83 and 5.766 values between 5.76 and 6.684 values between 6.68 and 7.601 values between 7.60 and 8.52offnegativepositiveoften

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. anke
  2. 嗓子
  3. 能力强
  4. 经验丰富
  5. wap
  6. HAVE
  7. 邗
  8. 可行性
  9. 这是一种
  10. türlü

Suppresses

  1. ments
  2. ment
  3. servername
  4. 建军
  5. by
  6. currently
  7. previously
  8. ging
  9. by
  10. rom