Neuron Explorer

Qwen3-0.6B

comma after a month and day

A neuron in layer 6 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 lights up on the comma that comes right after a month and day in a date, such as 'January 11,' or 'April 2nd,'. When it was turned up, the text drifted into unrelated topics, so the steering results don't show a clear effect. Its strongest opposite reaction is on the word 'mid' in phrases like 'mid-December'.

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

    ovember 5, 1818 – January 11, 1893)⏎Major general of the Union Army, politician, lawyer

  2. #2 peak

    ⏎• 13-19 July, 2020 - Helsinki, Finland⏎• 20-2

  3. #3 peak

    then, the Civil War Center is committed to the following:⏎WHY KENNESAW STATE UNIVERSITY IS AN⏎<|endoftext|>[Last Updated: April 2nd, 2015]⏎Conway's Game of Life⏎Conway's

  4. #4 peak

    Science Grants by Hughes Institute⏎By WILLIAM CELIS 3d⏎Published: October 4, 1994⏎An $86 million grant to 62

  5. #5 peak

    Its Implications to our Behavior and History⏎Spring 2017 : Jan. 23rd - April 2nd, 2017 - 10 Week Online Course⏎Summer 2

  6. #6 peak

    in the same book, “The milpa is one of the most successful human inventions ever created.”⏎© 2009 – 2020, Rudy Giron. All rights reserved.<|endoftext|>What condition do Neil Diamond and Billy

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:What's thebest way touse aweb app toget aquestion with aquestionable question ofaquestionable topic? Okay,sotheuser wants toget aquestion aboutaquestionable topic,butthequestion isnotclear.

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

    John Adams left his farm in Massachusetts for the Continental Congress in Philadelphia. Doesn’t sound particularly interesting, I know but consider this.⏎John Adams left his home in mid-December 1775 to<|endoftext|>For whatever reason, a lot of

  2. #2 peak

    real aircraft.⏎The D-21, constructed without landing gears, was an innovative contraption that could only be used only once and was jettisoned mid-air from another plane that acted as a mothership. It was the product of

  3. #3 peak

    Germans had time to move in and reassert control over the camps.⏎An operation to rescue and evacuate personnel in the Marche and Abruzzi regions was requested in mid September by 15 Army Group, following concerns that the Germans would take steps

  4. #4 peak

    pumped through their tanks and they feed on nutrients in the water. For the first batch of oysters, which are spawned in February, they reach this size in mid-March.⏎The oysters are eventually transferred to mesh bags when they are an

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. L5:422 suppresses
  2. L5:667 suppresses
  3. L4:874 weakly suppresses opening parenthesis before acronyms
  4. L4:2129 weakly suppresses
  5. L5:9 weakly suppresses
  6. L4:2343 weakly excites
  7. L4:2830 weakly excites
  8. L1:491 weakly excites
  9. L1:547 weakly suppresses
  10. L0:3008 weakly excites
  11. L0:1340 weakly excites
  12. L2:766 weakly excites
  13. L1:1857 weakly suppresses
  14. L0:1053 weakly suppresses
  15. L1:2775 weakly excites
  16. L4:1433 weakly suppresses
  17. L4:1769 weakly suppresses
  18. L1:635 weakly excites 'by' after a past participle
  19. L2:2328 weakly suppresses
  20. L2:1401 weakly suppresses
  21. L2:778 weakly suppresses
  22. L4:1631 weakly excites
  23. L5:3006 weakly excites
  24. L4:47 weakly suppresses

Feeds later layers

  1. L14:140 strongly suppresses
  2. L16:396 strongly suppresses
  3. L16:1421 strongly suppresses
  4. L16:723 strongly excites
  5. L18:76 suppresses blank spot before a number
  6. L27:33 suppresses
  7. L17:15 excites
  8. L27:11 excites unclear: mixed fragments in lists and pronunciations
  9. L26:21 excites
  10. L20:59 suppresses
  11. L20:489 suppresses
  12. L24:51 suppresses
  13. L22:16 excites comma in dates and birth lines
  14. L18:91 suppresses
  15. L19:509 suppresses
  16. L25:30 suppresses
  17. L24:39 suppresses
  18. L27:126 suppresses
  19. L20:433 excites
  20. L23:80 suppresses blank slot before a year or number
  21. L26:32 excites
  22. L14:309 excites
  23. L21:29 suppresses
  24. L22:14 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 -2.02 and -1.941 values between -1.39 and -1.311 values between -1.31 and -1.241 values between -1.24 and -1.162 values between -1.16 and -1.081 values between -1.08 and -1.002 values between -0.92 and -0.851 values between -0.85 and -0.772 values between -0.77 and -0.691 values between -0.69 and -0.616 values between -0.61 and -0.535 values between -0.53 and -0.4513 values between -0.45 and -0.3823 values between -0.38 and -0.3051 values between -0.30 and -0.22153 values between -0.22 and -0.14398 values between -0.14 and -0.061568 values between -0.06 and 0.021026 values between 0.02 and 0.09393 values between 0.09 and 0.17166 values between 0.17 and 0.2552 values between 0.25 and 0.3316 values between 0.33 and 0.417 values between 0.41 and 0.487 values between 0.48 and 0.564 values between 0.56 and 0.643 values between 0.64 and 0.721 values between 0.72 and 0.801 values between 0.88 and 0.951 values between 0.95 and 1.031 values between 1.03 and 1.11offnegativepositiveoften

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

  2. as
  3. in
  4. a
  5. "
  6. to
  7. In
  8. us
  9. V
  10. n

Suppresses

  1. isphere
  2. iedad
  3. shaw
  4. tré
  5. ancia
  6. uated
  7. ыта
  8. uencia
  9. istique
  10. irá