Neuron Explorer

Qwen3-0.6B

the space before a rounded number

A neuron in layer 19 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 blank space just before a number that follows words like 'about', 'approximately', 'an estimated' or 'more than'. It is a small, specific pattern. Its steered outputs were unrelated, so they don't help explain it.

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

    world? About 3 billion people earn less than two dollars a day, but figures for the rest

  2. #2 peak

    approximately 24 hours after laparoscopic intervention, due to the residual CO2 gas

  3. #3 peak

    at about 64,000 miles per hour, hitting at Sutter's Mill

  4. #4 peak

    loprid and clothianidin, produced by Bayer, and thiamethoxam, produced by Syngenta.⏎There are an estimated 2.62 million bee colonies in the United States.⏎-Noel Brinker

  5. #5 peak

    and around 935, rebels defeated it and founded the Koryo dynasty. This

  6. #6 peak

    017).⏎Approximately 65,000 individuals in the U.S. died in 2

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 question of whether the answer to a question is to be determined by the answer to the question of a question of a question... a question of a question of a question... question of a question of a question... Question of a question of

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

    turn anti-clockwise and then forward 3. Did you find any treasure?⏎8. 3/4 turn anti-clockwise, back 1, 1/4 turn anti-clockwise, forward

  2. #2 peak

    110.00000000.00000000 IP: 10101100.0001000x.xxxxxxxx.xxxxxxxx Each subnet

  3. #3 peak

    13⏎Updated: March 13, 2013⏎Updated: October 02, 2013⏎At the outset, it must be stated quite equivoc

  4. #4 peak

    solutions of A X = Y 1 . Solution : X = V and X = V + Z f) Find another solution of A X = 3 Y 1- 5 Y 2 . Solution : X = 3 V

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. L18:664 excites
  2. L18:439 excites
  3. L18:307 suppresses
  4. L18:2559 excites
  5. L18:2507 suppresses
  6. L18:1871 excites
  7. L18:548 excites
  8. L16:646 suppresses
  9. L18:761 excites
  10. L10:2348 weakly excites
  11. L18:140 weakly excites
  12. L3:115 weakly excites
  13. L18:152 weakly suppresses
  14. L12:539 weakly suppresses
  15. L15:432 weakly excites
  16. L18:1867 weakly suppresses
  17. L17:1051 weakly excites
  18. L17:2264 weakly excites
  19. L17:2271 weakly excites
  20. L13:855 weakly suppresses
  21. L16:1741 weakly suppresses
  22. L17:3066 weakly suppresses
  23. L17:795 weakly excites
  24. L18:1523 weakly excites

Feeds later layers

  1. L20:294 suppresses
  2. L24:363 suppresses
  3. L20:2065 suppresses
  4. L24:367 excites
  5. L21:12 excites
  6. L25:409 suppresses
  7. L20:524 excites
  8. L27:910 excites
  9. L26:732 weakly suppresses
  10. L25:5 weakly excites
  11. L21:224 weakly excites
  12. L27:625 weakly excites
  13. L22:225 weakly excites
  14. L20:1781 weakly excites
  15. L21:457 weakly excites
  16. L20:329 weakly excites
  17. L27:352 weakly excites
  18. L24:21 weakly excites
  19. L21:51 weakly excites
  20. L26:70 weakly suppresses
  21. L23:41 weakly excites 'at' after a verb or adjective
  22. L20:1189 weakly suppresses
  23. L27:289 weakly excites
  24. L26:315 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.

3 values between -3.91 and -3.561 values between -3.56 and -3.201 values between -2.85 and -2.496 values between -2.49 and -2.145 values between -2.14 and -1.7810 values between -1.78 and -1.4316 values between -1.43 and -1.0742 values between -1.07 and -0.72169 values between -0.72 and -0.361073 values between -0.36 and -0.001328 values between -0.00 and 0.35378 values between 0.35 and 0.71237 values between 0.71 and 1.06156 values between 1.06 and 1.42115 values between 1.42 and 1.7785 values between 1.77 and 2.1361 values between 2.13 and 2.4848 values between 2.48 and 2.8431 values between 2.84 and 3.2031 values between 3.20 and 3.5524 values between 3.55 and 3.9118 values between 3.91 and 4.2614 values between 4.26 and 4.6212 values between 4.62 and 4.9711 values between 4.97 and 5.3311 values between 5.33 and 5.683 values between 5.68 and 6.043 values between 6.04 and 6.403 values between 6.40 and 6.752 values between 6.75 and 7.113 values between 7.11 and 7.463 values between 7.46 and 7.821 values between 7.82 and 8.171 values between 8.17 and 8.531 values between 8.53 and 8.882 values between 9.95 and 10.31offnegativepositiveoften

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. 发达
  2. -first
  3. prime
  4. رين
  5. perí
  6. BASE
  7. além
  8. abez
  9. segundo
  10. cancellationToken

Suppresses

  1. '..
  2. onn
  3. 干活
  4. onis
  5. 带着
  6. bans
  7. ivic
  8. ..."⏎⏎
  9. 的因素
  10. anism