Neuron Explorer

Qwen3-0.6B

middle pieces of split surnames

A neuron in layer 22 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 fires on a fragment in the middle of a person's surname or a name that the tokenizer has split into pieces, such as the 'L' in Lerdorf or Llosa. Turning it down made the model get stuck repeating the word 'with', while turning it up gave loose, unrelated sentences.

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

    the server.⏎- PHP was created by Rasmus Lerdorf in 1995.⏎- Though PHP was originally known as

  2. #2 peak

    bad soil” or “land without anything growing on it”.⏎Firstly, peculiarity of this place noticed in 1936 by V.Vahrameev and later in

  3. #3 peak

    been accused of being traitors because they left their homelands, from Dante to Joseph Conrad and Joyce, García Marquez and Gunther Grass and Vargas Llosa? And whatever the explanation that those posing the question—

  4. #4 peak

    writes, but rather evaluate him by where he lives, or by the location of "the room" from which he writes, as the Peruvian writer Vargas Llosa noted in his commentary on the issue in one of his articles.⏎This

  5. #5 peak

    support the case for a largely periodic, coordinated, and intermittently catastrophic geologic record, which is a departure from the views<|endoftext|>Dr. Folkert van Werven⏎Dr. Alexi Goranov⏎Cells, regardless of whether they

  6. #6 peak

    al scientists with a strong interest in teaching.⏎Folkert van Werven, and Alexi Goranov. 7.342 To

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: with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with with

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

    the sensation of vertigo. The condition occurs most often in people 50 and older and is more common in women than in men.⏎Doctors treat BPPV by

  2. #2 peak

    range often experiences greater exposure to bacteria (via school), leading to infection and tonsillitis. However, adults can also experience chronic tonsillitis which can be treated by receiving a tonsillectomy.⏎The surgical procedure for

  3. #3 peak

    sensors, buttons, alligator clip pads and more. Build projects with Circuit Playground in a few minutes with the drag-and-drop MakeCode programming site, learn computer science using the CS Discoveries class on code.org, jump into CircuitPython to learn

  4. #4 peak

    . The Cyberknife technology works so precisely that the administered dose only reaches the tumor and spares surrounding tissue, organs or nerves.⏎Which tumors and lesions can be treated with Cyberknife?⏎The Cyberknife System is designed for radiosurgical treatment of tumors

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. L19:2736 strongly excites
  2. L21:523 strongly excites
  3. L16:2312 excites
  4. L18:2599 suppresses
  5. L19:1689 excites
  6. L18:2460 excites
  7. L14:1330 excites
  8. L19:2321 suppresses
  9. L20:2484 suppresses
  10. L19:2940 excites
  11. L21:1488 suppresses
  12. L20:2516 suppresses
  13. L21:2223 suppresses "to" after "used" (purpose)
  14. L20:3044 excites
  15. L21:2595 excites
  16. L20:501 suppresses
  17. L20:1525 excites
  18. L18:1127 excites
  19. L18:2975 excites
  20. L18:2903 excites
  21. L10:430 suppresses
  22. L17:2474 suppresses
  23. L20:1758 excites
  24. L16:2321 suppresses

Feeds later layers

  1. L26:741 strongly suppresses
  2. L24:921 strongly suppresses
  3. L27:603 strongly suppresses
  4. L24:883 strongly suppresses
  5. L26:223 excites
  6. L23:984 suppresses
  7. L24:2419 excites
  8. L27:1443 suppresses
  9. L23:684 weakly excites
  10. L27:2153 weakly suppresses
  11. L23:1235 weakly suppresses
  12. L27:1606 weakly suppresses
  13. L27:1323 weakly suppresses
  14. L25:2139 weakly excites capital N starting an abbreviation
  15. L23:2461 weakly suppresses
  16. L24:2530 weakly excites
  17. L23:1292 weakly suppresses
  18. L25:1114 weakly suppresses
  19. L27:430 weakly suppresses
  20. L27:2572 weakly suppresses
  21. L23:1073 weakly excites
  22. L23:2303 weakly excites
  23. L24:1741 weakly excites
  24. L23:605 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 -13.72 and -13.261 values between -11.87 and -11.412 values between -9.10 and -8.633 values between -7.71 and -7.251 values between -7.25 and -6.783 values between -6.78 and -6.324 values between -5.86 and -5.402 values between -5.40 and -4.934 values between -4.93 and -4.4711 values between -4.47 and -4.019 values between -4.01 and -3.557 values between -3.55 and -3.0810 values between -3.08 and -2.6218 values between -2.62 and -2.1625 values between -2.16 and -1.7036 values between -1.70 and -1.2359 values between -1.23 and -0.77183 values between -0.77 and -0.311042 values between -0.31 and 0.151634 values between 0.15 and 0.62752 values between 0.62 and 1.0887 values between 1.08 and 1.5411 values between 1.54 and 2.002 values between 2.00 and 2.471 values between 4.32 and 4.78offnegativepositiveoften

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. izu
  2. osterone
  3. 一字
  4. edef
  5. ]|[
  6. ethyst
  7. achen
  8. ','=',$
  9. utherford
  10. ackle

Suppresses

  1. with
  2. with
  3. With
  4. With
  5. _with
  6. →with
  7. WITH
  8. với
  9. avec
  10. .with