Neuron Explorer

Qwen3-0.6B

pieces of words split into fragments

A neuron in layer 2 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a small piece in the middle of a longer word that the model has split into parts, such as 'Jant' in 'Jantzen' or 'at' in 'insatiable'. Mostly it shows up in names and uncommon words. Turning it up did not produce a clear theme in the written text.

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 gene activity among individuals can give powerful clues for diagnosis, prognosis and intervention of different diseases. Senior author Emmanouil Dermitzakis, Louis Jeantet Professor at UNIGE, who led the study, emphasizes that today’s

  2. #2 peak

    30,000 guilders. During this period David Penner was a minister of the church of Bagdannen, whereas Salomon Jantzen seems to have been a deacon. The members of this Lithuanian

  3. #3 peak

    . The Art of Slow Reading. Portsmouth, NH: Heinamann, 2012.⏎And advocate to taking one’s time to

  4. #4 peak

    23, 2011⏎When two galaxies merge to form a giant, the central supermassive black hole in the new galaxy develops an insatiable appetite. However, this ferocious appetite is unsustainable.⏎The growth of super

  5. #5 peak

    of 1934, the Franks r…⏎E-pasta adrese, uz kuru nosūtīt darba saiti:⏎Saite uz darbu:<|endoftext|>Branch of biology that groups and names organisms based on studies

  6. #6 peak

    dharma of another. Nothing is ever lost in following one’s own dharma. But competition in another’s dharma breeds fear and insecurity.”⏎The word janma translates as “birth”. Ashtami (ush-tum-me) is the

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 a number of records. I have a number of records. I have a number of records. I have a number of records. I have a number of records. I have a number of records. I have a number of records

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

    , pickled pigs feet, lutfisk or dried codfish and variety of sweets.⏎A popular Christmas tradition is to

  2. #2 peak

    unkel⏎starring Jim Witter,⏎featuring Ian Tanner⏎“Time it was, and what a time it

  3. #3 peak

    land.⏎The cut comes into effect on December 19 and will apply across England and Wales.⏎According to Defra over 70% of the nitrates in Britain’s waterways can be traced back to agriculture. They cause twin

  4. #4 peak

    water contaminants discovered by water testing labs.⏎Nitrates are routinely discovered in tainted water supplies because they’re so prevalent in the environment. The primary source of nitrates in the

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. L0:2257 suppresses
  2. L0:1301 weakly suppresses
  3. L0:2714 weakly suppresses
  4. L1:21 weakly excites
  5. L0:178 weakly suppresses
  6. L0:1224 weakly suppresses
  7. L1:2340 weakly suppresses
  8. L1:2469 weakly suppresses
  9. L1:2801 weakly excites
  10. L1:478 weakly excites
  11. L1:951 weakly excites
  12. L1:2505 weakly suppresses
  13. L1:874 weakly excites
  14. L1:1595 weakly suppresses
  15. L1:990 weakly suppresses
  16. L0:286 weakly excites
  17. L0:659 weakly suppresses
  18. L1:2683 weakly excites
  19. L0:2209 weakly excites
  20. L1:87 weakly suppresses
  21. L0:101 weakly suppresses
  22. L0:2919 weakly suppresses
  23. L1:2433 weakly suppresses
  24. L0:814 weakly suppresses

Feeds later layers

  1. L8:2323 weakly excites
  2. L4:230 weakly suppresses
  3. L10:2594 weakly excites
  4. L16:814 weakly suppresses
  5. L12:2283 weakly excites
  6. L17:81 weakly suppresses
  7. L24:2468 weakly suppresses
  8. L14:3021 weakly excites
  9. L15:1869 weakly excites
  10. L19:2525 weakly excites
  11. L12:467 weakly suppresses
  12. L14:605 weakly excites
  13. L6:1899 weakly excites
  14. L21:858 weakly excites
  15. L25:2353 weakly suppresses
  16. L20:321 weakly excites
  17. L24:2841 weakly excites
  18. L17:2965 weakly excites
  19. L19:673 weakly suppresses
  20. L9:1821 weakly suppresses
  21. L14:275 weakly excites
  22. L3:686 weakly excites
  23. L26:869 weakly excites word before him or he
  24. L13:921 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 -0.73 and -0.711 values between -0.68 and -0.651 values between -0.41 and -0.391 values between -0.39 and -0.364 values between -0.36 and -0.335 values between -0.33 and -0.302 values between -0.30 and -0.2722 values between -0.27 and -0.2430 values between -0.24 and -0.2144 values between -0.21 and -0.1892 values between -0.18 and -0.15123 values between -0.15 and -0.12174 values between -0.12 and -0.09302 values between -0.09 and -0.07424 values between -0.07 and -0.04550 values between -0.04 and -0.01604 values between -0.01 and 0.02461 values between 0.02 and 0.05381 values between 0.05 and 0.08230 values between 0.08 and 0.11193 values between 0.11 and 0.14102 values between 0.14 and 0.1773 values between 0.17 and 0.2038 values between 0.20 and 0.2320 values between 0.23 and 0.2616 values between 0.26 and 0.286 values between 0.28 and 0.314 values between 0.31 and 0.342 values between 0.34 and 0.371 values between 0.37 and 0.401 values between 0.40 and 0.43offnegativepositiveoften

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. erg
  2. onium
  3. one
  4. erto
  5. rom
  6. erge
  7. 级
  8. 一口气
  9. eu
  10. rise

Suppresses

  1. stash
  2. arness
  3. xffffffff
  4. ��
  5. Utah
  6. AFX
  7. .LookAndFeel
  8. chatt
  9. .hashCode
  10. ätz