Neuron Explorer

Qwen3-0.6B

commas in dates and number lists

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 lights up on commas that separate parts of a date, a year, or a list of numbers, such as 'November 18, 1918' or 'levels 1,2, and 3'. Turning it up made the model's writing turn into scrambled, repetitive text, so the effect is not easy to read.

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

    pass Level 2, they have to say level 1 and 2 correctly. To pass level 3, they have to say level 1,2, and 3 correctly. To pass level 4, the must say all of

  2. #2 peak

    20|⏎|Order Information:||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce

  3. #3 peak

    ovember 5, 1818 – January 11, 1893

  4. #4 peak

    100⏎tspan=[t0 tend]; %need to use it in this form in MATLAB 5.⏎[t,y]=ode45('rhs',tspan,y0)%call the routine ode45⏎plot(t,y)

  5. #5 peak

    The basic⏎source of all minerals is the hot magma in the⏎interior of the earth.⏎(more content follows the advertisement below) A D V E R T I S E M E N T⏎When magma cools,⏎

  6. #6 peak

    THE⏎For more recent information about the government, see Facts about Albania.⏎Source: U.S. Library of Congress<|endoftext|>» 2010.aug.25 » We have the Russians to thank for the Internet. In 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: The more than

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

    , and we know that,⏎Lines with the same slope are parallel.⏎Hence, the given lines are parallel.⏎Let's consider another set of equations.⏎y = 2x +3 and 4y = –2x+10

  2. #2 peak

    line, then they are perpendicular.⏎Let's see how to identify parallel and perpendicular lines.⏎Let's take some examples to understand this.⏎Considering two equations:⏎y = 2x +3 and y = 2x+5⏎On comparing

  3. #3 peak

    bigger than a two quarter⏎one quarter = two eighths therefore, three eighths is larger than one quarter⏎One quarter is 1/4 = 4/16 and three sixteenths is 3/1

  4. #4 peak

    , 24 = 4 ⋅ 6. So to solve the second part, on

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. L21:164 suppresses
  2. L21:214 excites
  3. L14:35 suppresses
  4. L20:2320 suppresses
  5. L20:1298 suppresses
  6. L21:173 excites
  7. L19:1207 weakly suppresses
  8. L4:2864 weakly suppresses
  9. L19:534 weakly excites
  10. L19:726 weakly suppresses comma inside large numbers
  11. L18:2740 weakly suppresses
  12. L17:2940 weakly suppresses
  13. L5:539 weakly excites
  14. L17:118 weakly excites comma after a clause ending
  15. L18:115 weakly suppresses
  16. L19:220 weakly excites question mark ending a web question heading
  17. L14:309 weakly suppresses
  18. L20:966 weakly excites
  19. L17:215 weakly suppresses
  20. L19:1697 weakly suppresses
  21. L18:946 weakly suppresses
  22. L10:335 weakly excites commas in catalog, date, and list metadata
  23. L17:132 weakly suppresses
  24. L13:2204 weakly excites

Feeds later layers

  1. L24:51 strongly excites
  2. L26:17 strongly excites
  3. L23:239 strongly excites
  4. L25:222 excites
  5. L25:304 excites
  6. L25:146 excites
  7. L27:71 suppresses
  8. L23:262 excites
  9. L26:247 excites
  10. L25:109 suppresses
  11. L24:141 excites
  12. L26:103 excites blank space before a number
  13. L27:431 excites
  14. L26:224 excites
  15. L26:263 excites
  16. L25:217 suppresses
  17. L27:323 excites
  18. L27:513 excites
  19. L25:239 excites
  20. L27:514 excites
  21. L27:11 suppresses unclear: mixed fragments in lists and pronunciations
  22. L26:240 excites
  23. L27:220 excites end-of-text token between documents
  24. L26:199 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 -18.61 and -18.021 values between -14.46 and -13.872 values between -6.17 and -5.581 values between -5.58 and -4.992 values between -4.99 and -4.396 values between -4.39 and -3.8010 values between -3.80 and -3.2112 values between -3.21 and -2.6225 values between -2.62 and -2.0248 values between -2.02 and -1.4354 values between -1.43 and -0.84328 values between -0.84 and -0.252883 values between -0.25 and 0.35505 values between 0.35 and 0.9418 values between 0.94 and 1.535 values between 1.53 and 2.122 values between 2.12 and 2.722 values between 2.72 and 3.311 values between 3.31 and 3.902 values between 4.49 and 5.09offnegativepositiveoften

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. cre
  2. gl
  3. no
  4. standing
  5. She
  6. yield
  7. ine
  8. 결
  9. cum
  10. Indeed

Suppresses

  1. selected
  2. ;display
  3. "','
  4. !
  5. "','"
  6. !");⏎
  7. ','=',
  8. LSB
  9. !!
  10. powered