Neuron Explorer

Qwen3-0.6B

spot where a number or statistic comes next

A neuron in layer 24 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 tends to light up on the word just before a number, such as 'nearly', 'close to' or 'Issue No', so it seems to be getting ready for a figure. It is mixed, and when turned up the model mostly got stuck repeating words like 'the president' rather than writing anything about numbers.

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

    in a math book. What about the problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 13.09.2016 19:19

  2. #2 peak

    infections.⏎Sub-Saharan Africa remains most severely affected, with nearly 1 in every 25 adults living with HIV and accounting for nearly 70% of the people living with HIV worldwide.<|endoftext|>Everyone’s experienced a

  3. #3 peak

    18, July 6). 1929 Stock Market Crash and the Great Depression –⏎Documentary [Video File]. Retrieved from https://youtu.be/qlSxPouPCIM⏎The post 1929

  4. #4 peak

    and in some countries the majority of children are not vaccinated. In 2015, around 20 million babies missed their measles shots and an estimated 134,000 children died from the disease.⏎The Democratic Republic

  5. #5 peak

    localised muscular contractions. Finally, generalised spasms may occur, often leading to death from heart and lung failure. The overall death rate is close to 50%, depending on the clinical presentation, patient’s age and medical support.

  6. #6 peak

    .⏎Posted 31 Aug 2009 • 1 answer⏎Posted 20 Jan 2010 • 1 answer⏎Posted 27 Jun 2010 • 1 answer⏎Posted 1

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: A single person in a single room with a single lamp that is on all the time could go on a long trip all over the world all the right right right right right right ... all right right right right right right ... maybe on the t

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

    58 KJV⏎• Luke 9:58 Commentaries⏎• Luke 9:58 Bible Apps⏎• Luke 9:58 Biblia Paralela⏎• Luke 9:58 Chinese Bible⏎•

  2. #2 peak

    're gone.⏎The title of this well-known story comes from a poem by Sara Teasdale in which nature is depicted as indifferent to mankind's absence after its self

  3. #3 peak

    , Ghazipur,Bhagalpur (towns)⏎Step 3 Through the Rajmahal Hills⏎Step 4 Mirzapur, Ballia, Buxar, Saidpur and Chunar -(More towns)⏎Step 5 The

  4. #4 peak

    th centuries when epidemics of yellow fever and cholera commonly killed thousands and thousands of people. This fear was not totally unfounded. An Englishman, William Tebb, did a study of premature

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. L10:220 suppresses hyphen inside compound words
  2. L14:102 weakly excites
  3. L18:2497 weakly excites hyphen inside compound words
  4. L16:222 weakly excites
  5. L19:586 weakly excites
  6. L8:7 weakly suppresses
  7. L20:124 weakly suppresses
  8. L5:608 weakly excites
  9. L0:1449 weakly excites
  10. L17:550 weakly excites
  11. L14:140 weakly suppresses
  12. L18:569 weakly suppresses
  13. L6:715 weakly excites
  14. L5:28 weakly suppresses
  15. L2:258 weakly excites
  16. L17:1125 weakly excites
  17. L18:377 weakly excites number right after 'about' or 'less than'
  18. L9:2403 weakly suppresses
  19. L1:658 weakly suppresses
  20. L22:2496 weakly suppresses
  21. L2:28 weakly suppresses
  22. L17:868 weakly excites
  23. L18:269 weakly excites
  24. L21:360 weakly suppresses 'such' introducing examples

Feeds later layers

  1. L26:667 weakly suppresses
  2. L25:295 weakly suppresses
  3. L25:2759 weakly suppresses
  4. L27:886 weakly suppresses
  5. L25:24 weakly suppresses
  6. L25:710 weakly suppresses
  7. L25:146 weakly suppresses
  8. L26:1280 weakly excites
  9. L27:2855 weakly suppresses
  10. L25:879 weakly suppresses
  11. L27:1507 weakly excites
  12. L25:184 weakly suppresses
  13. L27:144 weakly suppresses
  14. L25:2295 weakly suppresses
  15. L27:101 weakly excites
  16. L27:1296 weakly suppresses
  17. L25:3002 weakly suppresses
  18. L27:962 weakly suppresses
  19. L26:1171 weakly excites
  20. L27:21 weakly excites
  21. L27:85 weakly excites
  22. L27:796 weakly suppresses
  23. L25:55 weakly suppresses
  24. L25:1718 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 -33.66 and -31.721 values between -25.92 and -23.992 values between -23.99 and -22.062 values between -20.12 and -18.193 values between -18.19 and -16.253 values between -16.25 and -14.325 values between -14.32 and -12.395 values between -12.39 and -10.457 values between -10.45 and -8.5218 values between -8.52 and -6.5916 values between -6.59 and -4.6530 values between -4.65 and -2.72119 values between -2.72 and -0.792286 values between -0.79 and 1.15738 values between 1.15 and 3.08277 values between 3.08 and 5.02137 values between 5.02 and 6.9571 values between 6.95 and 8.8853 values between 8.88 and 10.8239 values between 10.82 and 12.7526 values between 12.75 and 14.6814 values between 14.68 and 16.6211 values between 16.62 and 18.556 values between 18.55 and 20.488 values between 20.48 and 22.425 values between 22.42 and 24.356 values between 24.35 and 26.293 values between 26.29 and 28.223 values between 28.22 and 30.152 values between 30.15 and 32.094 values between 32.09 and 34.021 values between 34.02 and 35.952 values between 35.95 and 37.891 values between 37.89 and 39.821 values between 39.82 and 41.752 values between 41.75 and 43.69offnegativepositiveoften

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. separator
  2. 乜
  3. 隔壁
  4. locals
  5. pwd
  6. parallel
  7. 接地气
  8. stud
  9. tea
  10. IMPLEMENT

Suppresses

  1. ać
  2. ków
  3. ::⏎
  4. uisse
  5. %⏎⏎
  6. quotation
  7. xaa
  8. THON
  9. BaseController
  10. Flem