Neuron Explorer

Qwen3-0.6B

opening year or date range

A neuron in layer 22 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 space right before a year or date, such as the years in a historical dating phrase. Turning it up gave wandering, unrelated text, and turning it down made the model repeat words like can't or is.

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

    1185 CE), Japanese art flourished, and new forms of art, such as screen painting, began to emerge. The Kamakura period (1185 CE to 1333 CE) saw the rise

  2. #2 peak

    galaxy of ours — and the galaxies beyond. And we have Copernicus, among others, to thank for that.⏎Copernicus was born in Poland in 1473 and as a young man became fascinated by math, medicine and

  3. #3 peak

    the River May (now St. John’s, Florida), in 1564. […]⏎Jacques le Moyne de Morgues (c. 1533–1588) was a French artist and member

  4. #4 peak

    He loved them very much. They always used to help one another<|endoftext|>Ramses III, Ramses also spelled Ramesses or Rameses (died 1156 bce, Thebes, Egypt), king of ancient Egypt

  5. #5 peak

    revised, January 2003, December 2006⏎Part I: The Umayyad Caliphate (661 – 750 CE)⏎5 Tibet on the Eve of the Arrival of the First

  6. #6 peak

    ⏎Ink painting originated in China during the Tang dynasty (AD 618-907) and was known in Japan by the eighth

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 can't read my own record, but I can read the record of my fellow Christians. Can I do that? Can my record, as a fellow Christian, have the record of my fellow Christians? Can I hold my record as a

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

    06:37. Five midget submarines had been assigned to torpedo U.S. ships after the bombing started. None of these returned, and only four have since been found. Of the ten sailors aboard, nine died; the only survivor

  2. #2 peak

    24, 1896 – December 21,1940) was an Irish American Jazz Age novelist and short story writer, who is generally regarded as one of the greatest American writers of the Twentieth Century. Fitzgerald

  3. #3 peak

    eventually to be added to Olympic National Park.⏎Besides serving as postmaster, Fannie operated the store and overnight accommodations at Mora. A few of her diaries, written from 1914 to 1922, have

  4. #4 peak

    0, recounts how Radbod's son was baptised and "freed from the flesh", and died soon after. Unfortunately, no name of the son is given. Was he Fris perhaps?⏎The story of Fris reminds us of

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. L17:2618 suppresses
  2. L18:2893 suppresses
  3. L17:2799 suppresses
  4. L14:2753 excites
  5. L10:795 suppresses
  6. L16:1532 excites
  7. L20:2032 excites
  8. L20:1296 excites
  9. L12:1855 excites
  10. L10:781 excites
  11. L17:1631 excites
  12. L17:1767 excites
  13. L6:2771 excites
  14. L15:1138 weakly excites
  15. L19:1551 weakly suppresses
  16. L18:535 weakly suppresses
  17. L11:2775 weakly suppresses digits inside years like 19xx
  18. L18:2489 weakly suppresses
  19. L19:2536 weakly excites
  20. L13:566 weakly excites
  21. L21:2326 weakly suppresses
  22. L13:2491 weakly suppresses
  23. L9:508 weakly suppresses
  24. L20:388 weakly suppresses

Feeds later layers

  1. L23:535 strongly excites
  2. L24:60 strongly suppresses
  3. L24:320 strongly excites
  4. L27:786 strongly excites
  5. L25:848 strongly excites
  6. L23:108 strongly suppresses
  7. L24:252 excites
  8. L27:382 excites
  9. L24:778 excites
  10. L25:1841 excites
  11. L24:780 excites subject pronouns inside subordinate clauses
  12. L25:1726 excites
  13. L27:565 suppresses
  14. L25:1711 excites
  15. L27:1256 excites
  16. L24:277 suppresses
  17. L24:1370 excites
  18. L27:321 suppresses
  19. L23:257 excites
  20. L26:1716 excites
  21. L25:2693 weakly excites
  22. L24:258 weakly suppresses
  23. L26:1113 weakly excites
  24. L25:933 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.

2 values between -8.48 and -7.371 values between -7.37 and -6.269 values between -6.26 and -5.167 values between -5.16 and -4.0518 values between -4.05 and -2.9436 values between -2.94 and -1.84155 values between -1.84 and -0.733139 values between -0.73 and 0.38404 values between 0.38 and 1.4884 values between 1.48 and 2.5917 values between 2.59 and 3.6915 values between 3.69 and 4.807 values between 4.80 and 5.913 values between 5.91 and 7.012 values between 7.01 and 8.123 values between 8.12 and 9.231 values between 9.23 and 10.334 values between 10.33 and 11.441 values between 34.68 and 35.78offnegativepositiveoften

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. Was
  2. was
  3. Was
  4. was
  5. wasn
  6. were
  7. Were
  8. Were
  9. _was
  10. weren

Suppresses

  1. can
  2. ister
  3. isters
  4. accounts
  5. cannot
  6. izes
  7. encv
  8. ISTER
  9. ustr
  10. will