Neuron Explorer

Qwen3-0.6B

period after month or initial

A neuron in layer 16 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 full stop that follows a short abbreviation, like the dot in 'Aug.' or after a person's initial such as 'F. Scott'. It is quiet on the full stop that closes a quotation or bracket. Turning it up or down gave text that did not show a clear link to this pattern.

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

    making glass and an ever-expanding range of uses, American-made glass has become a great source of interest for historical researchers and collectors alike.⏎On Saturday, Jan. 10, at the Coventry Community Center,

  2. #2 peak

    ID also called on Chile to speed up its judicial process and create a national plan to search for disappeared persons.⏎From Jurist, Aug. 23. Used with permission.<|endoftext|>Oise River, river that rises

  3. #3 peak

    that pheromones – in contrast to chemical pesticides – have no adverse effects on animals or humans.<|endoftext|>F. Scott Fitzgerald⏎Francis Scott Key Fitzgerald (September 24, 1

  4. #4 peak

    to perform basic statistical functions using both a scientific calculator and Microsoft Excel.⏎Saturday, Aug. 23, 2-4:30 p.m.⏎Basic Statistics

  5. #5 peak

    person, except when referring to Hana, and without using quotation marks. Since there are no quotation marks, Ondaatje is using language as<|endoftext|>A.N.L. - folds (1996), for computer generated tape is a

  6. #6 peak

    through our internet radio service in Model X. Because AM audio quality can be very poor, we offer internet radio to give<|endoftext|>A New Brunswick Ghost Story⏎S. E. Schlosser⏎It was a cold winter and the snow was deep

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: 2.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

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

    control of bleeding following mild injury while preventing inappropriate intravascular thrombosis. (See "Overview of hemostasis".)<|endoftext|>How to save your life in a fire or other emergency⏎For those

  2. #2 peak

    light(s). It turns out, perhaps somewhat surprisingly, that there is a simple and intuitive

  3. #3 peak

    War on Terror’. Curiously, despite their proliferation, they have received relatively little academic attention. In

  4. #4 peak

    in hard copy. Check the Northern Research Station web site to request a printed copy of this publication.⏎- Our on-line publications are scanned and captured using Adobe Acrobat. During the capture process some typographical errors may occur. Please contact Sharon Hobrl

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. L4:7 strongly excites
  2. L2:9 strongly suppresses sentence-final periods vs dots in code
  3. L15:313 strongly excites decimal point inside numbers
  4. L5:168 suppresses
  5. L14:422 excites periods after initials and abbreviations
  6. L11:132 excites period ending an opening topic sentence
  7. L1:130 suppresses
  8. L3:595 excites
  9. L5:155 suppresses
  10. L8:143 suppresses
  11. L0:36 excites
  12. L5:1304 excites
  13. L15:241 excites
  14. L4:295 excites
  15. L15:123 weakly suppresses
  16. L0:348 weakly excites space token right after a period
  17. L14:51 weakly excites colon that introduces a statement
  18. L5:574 weakly excites
  19. L3:1555 weakly suppresses
  20. L8:16 weakly suppresses
  21. L6:305 weakly suppresses
  22. L9:16 weakly suppresses unclear: periods inside numbers and names
  23. L0:2838 weakly suppresses
  24. L0:3046 weakly excites

Feeds later layers

  1. L17:642 strongly suppresses
  2. L19:831 strongly suppresses
  3. L17:1218 strongly excites
  4. L18:713 strongly suppresses
  5. L20:505 strongly suppresses
  6. L17:347 strongly suppresses full stops ending informative sentences
  7. L18:1437 strongly suppresses
  8. L18:791 strongly suppresses
  9. L17:2084 excites
  10. L17:1772 suppresses
  11. L20:2301 suppresses
  12. L19:344 suppresses periods ending explanations
  13. L17:990 suppresses
  14. L21:1086 suppresses end of sentence before a claim or statistic
  15. L20:962 suppresses decimal points and dots in numbers
  16. L17:1155 suppresses opening parenthesis after a sentence
  17. L23:179 suppresses
  18. L27:145 suppresses dots inside web addresses
  19. L19:902 excites
  20. L24:353 suppresses decimal point inside a number
  21. L22:67 suppresses
  22. L19:2497 suppresses
  23. L22:282 suppresses period then line break in lists
  24. L21:524 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.

2 values between -16.25 and -15.733 values between -15.73 and -15.212 values between -15.21 and -14.693 values between -14.16 and -13.643 values between -13.64 and -13.126 values between -13.12 and -12.605 values between -12.60 and -12.088 values between -12.08 and -11.566 values between -11.56 and -11.0311 values between -11.03 and -10.518 values between -10.51 and -9.994 values between -9.99 and -9.477 values between -9.47 and -8.955 values between -8.95 and -8.438 values between -8.43 and -7.918 values between -7.91 and -7.387 values between -7.38 and -6.864 values between -6.86 and -6.346 values between -6.34 and -5.824 values between -5.82 and -5.303 values between -5.30 and -4.785 values between -4.78 and -4.254 values between -4.25 and -3.731 values between -3.73 and -3.211 values between -3.21 and -2.696 values between -2.69 and -2.174 values between -2.17 and -1.6517 values between -1.65 and -1.1223 values between -1.12 and -0.6036 values between -0.60 and -0.081826 values between -0.08 and 0.441531 values between 0.44 and 0.96326 values between 0.96 and 1.4813 values between 1.48 and 2.011 values between 2.01 and 2.531 values between 4.09 and 4.61offnegativepositiveoften

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. alte
  2. inary
  3. ENTITY
  4. olated
  5. 椰
  6. =com
  7. AEA
  8. Ζ
  9. ват
  10. ungal

Suppresses

  1. plier
  2. гин
  3. 一个
  4. 打法
  5. ento
  6. ään
  7. しかも
  8. At
  9. ора
  10. -note