Neuron Explorer

Qwen3-0.6B

unclear: document ends and answer colons

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 is hard to pin down. It fires most strongly at the invisible marker that separates one web page from the next, and it switches off hardest on the colon in 'Answer:'. Turning it up made the model ramble and repeat phrases, which does not show a clear theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    code<|endoftext|>Jet engines, or gas turbines, suck in air at the front with a fan

  2. #2 peak

    wrong<|endoftext|>External Web sites⏎Britannica Web sites⏎Articles from Britannica encyc

  3. #3 peak

    groups<|endoftext|>What happens when you mix vinegar and baking powder?⏎When baking soda is mixed with

  4. #4 peak

    officers<|endoftext|>World Science Day for Peace and Development 2019⏎Open Science,

  5. #5 peak

    discarding of<|endoftext|>from The American Heritage® Dictionary of the English Language, 4th Edition⏎

  6. #6 peak

    9<|endoftext|>Virtual Kid Quest Guide for an Outdoor Art Quest⏎In celebration of Earth Day,

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: In the case of the long term, the best of the world for the world's future to "be well with us under a name of a name" (as we often do) and we "to be better for us under 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

    70%.⏎Answer: B. Humidity levels should be kept below 60 percent, ideally between

  2. #2 peak

    are parallel, perpendicular or neither?⏎Parallel lines are the lines that never intersect and perpendicular lines are the lines that intersect at 90 degrees.⏎Answer: Lines with the same slope are parallel and if the slope of one line is the

  3. #3 peak

    done.⏎Answer: True. There are hundreds of thousands of dust mites in our homes, from

  4. #4 peak

    i.e., "I don't want to see it again").⏎In #2 "saw" is past tense and neutral about the present. (see The<|endoftext|>In the field of meta-ethics, scholars have identified key problems 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. L9:1130 strongly suppresses
  2. L10:1713 suppresses
  3. L5:2946 excites
  4. L3:163 excites
  5. L5:334 excites
  6. L9:1757 suppresses
  7. L6:1228 weakly excites
  8. L4:1521 weakly excites
  9. L9:511 weakly excites
  10. L10:330 weakly excites
  11. L11:2445 weakly suppresses
  12. L3:609 weakly excites
  13. L0:683 weakly excites
  14. L6:2746 weakly suppresses
  15. L11:1214 weakly suppresses
  16. L6:840 weakly suppresses
  17. L2:1869 weakly excites
  18. L11:723 weakly suppresses
  19. L3:1898 weakly excites
  20. L3:2984 weakly suppresses unclear: mid-word subword pieces
  21. L6:228 weakly suppresses comma after a filler word
  22. L3:2704 weakly suppresses
  23. L4:1396 weakly excites "make" or "do" before an object noun
  24. L4:2015 weakly excites

Feeds later layers

  1. L18:461 suppresses
  2. L17:1340 excites
  3. L21:282 suppresses
  4. L17:816 suppresses
  5. L23:200 suppresses
  6. L17:244 excites
  7. L18:1382 weakly suppresses
  8. L21:1042 weakly excites
  9. L17:1680 weakly suppresses
  10. L18:191 weakly excites
  11. L21:1856 weakly suppresses
  12. L18:581 weakly suppresses
  13. L18:589 weakly excites
  14. L25:928 weakly excites
  15. L21:1766 weakly suppresses 'in' in 'bear in mind' phrases
  16. L19:327 weakly suppresses
  17. L18:1512 weakly excites end-of-document boundary token
  18. L18:2028 weakly suppresses
  19. L18:632 weakly excites
  20. L25:113 weakly suppresses
  21. L19:1000 weakly suppresses
  22. L27:2551 weakly excites
  23. L20:1072 weakly suppresses
  24. L18:2566 weakly 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 -11.12 and -10.821 values between -8.43 and -8.132 values between -5.44 and -5.141 values between -5.14 and -4.843 values between -3.95 and -3.651 values between -3.65 and -3.351 values between -3.35 and -3.051 values between -2.75 and -2.451 values between -2.45 and -2.155 values between -2.15 and -1.854 values between -1.85 and -1.559 values between -1.55 and -1.2611 values between -1.26 and -0.9611 values between -0.96 and -0.6624 values between -0.66 and -0.36275 values between -0.36 and -0.062907 values between -0.06 and 0.24618 values between 0.24 and 0.5432 values between 0.54 and 0.84offnegativepositiveoften

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. anship
  2. ,...⏎
  3. ,…
  4. ',//
  5. vows
  6. div
  7. iration
  8. inspirational
  9. ,...⏎⏎
  10. ','');⏎

Suppresses

  1. thems
  2. ptime
  3. cé
  4. PJ
  5. ij
  6. Buenos
  7. ("\\
  8. ["
  9. кри
  10. oma