Neuron Explorer

Qwen3-0.6B

"of" after all three, both, ten

A neuron in layer 5 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 word "of" when it comes right after a counting word like "all three", "both" or "ten", as in "all three of these". When it was turned up, the model's writing became repetitive and didn't pick up any clear new theme.

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

    radical reforms and a long period of growth. A year later, Ronald Reagan was elected President of the United States, and also embraced free market policies. All three of these leaders professed inspiration from the work of Milton Friedman. It is natural,

  2. #2 peak

    Up until the 1950s, the Folger also housed all nine of Gregory’s plaster reliefs in storage. It wasn’t until Dr. Frederick Hard

  3. #3 peak

    you.⏎It’s easy to use a converter for all types of users. You can even gift it to your kids so that they can learn the differences between both of these numerals.<|endoftext|>ventilation, extents and position of obstacles etc., that

  4. #4 peak

    5:45 PM #1⏎FNG methods of attack question⏎I read this a couple times and I'm still confused. I was taught all three of these as interior attack methods based on fire

  5. #5 peak

    its behavior when a mechanical force is exerted on it. Rheology is distinguished from fluid dynamics (see fluid mechanics) in that it is concerned with all three of the traditional states of matter rather than only with liquids and gases. Unlike polymer physics

  6. #6 peak

    measures to ensure that the process of arbitration was carried out properly. All of 13th century Christian Europe willingly looked upon him as an international judge.⏎

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 like of a device. The following is the information of the problem statement. So, for example, the problem of 9.2.18.2.3. In this case, the problem of 9.2.

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

    631–642, 2013. My thoughts precipitated a brief conversation with Erella Hovers in which, for the umpteenth time, she expounded the theory that a fossil

  2. #2 peak

    they were, literally.⏎Geographically they were as well, since they originated from at⏎least six distinct nationalities within the vast Russian Empire.⏎In every other respect, however, they were German, sharing a common⏎German culture and language.⏎

  3. #3 peak

    turmoil in a person's physical, psychological, and social health. Eating disorders most commonly affect women and occur in adolescence and young adulthood, though they can occur in anyone, of any gender, at any age.⏎Eating disorders most commonly affect women

  4. #4 peak

    growing challenge to expert knowledge and learning piloted by experts.⏎Two questions spring to mind:⏎Boundaries to schooling⏎The dissolution of boundaries in education was a hotly disputed subject. Some welcomed the dissolution as a promise of change. Others challenged its

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. L4:1743 strongly excites
  2. L4:2473 strongly suppresses
  3. L4:285 strongly suppresses
  4. L1:818 excites
  5. L3:1249 suppresses
  6. L0:1824 suppresses
  7. L4:951 excites
  8. L1:518 suppresses
  9. L3:1456 excites
  10. L1:401 suppresses
  11. L3:1717 suppresses
  12. L4:731 suppresses
  13. L1:289 excites
  14. L4:52 suppresses
  15. L4:871 excites
  16. L4:137 suppresses
  17. L4:1032 suppresses
  18. L4:521 suppresses
  19. L4:2743 suppresses
  20. L1:844 suppresses
  21. L4:2335 suppresses
  22. L4:128 excites
  23. L4:968 suppresses
  24. L4:2589 excites

Feeds later layers

  1. L10:167 strongly excites
  2. L8:252 strongly suppresses 'of' after 'each' or 'one' or 'all'
  3. L18:602 strongly excites
  4. L17:985 strongly excites
  5. L6:84 strongly excites
  6. L7:173 strongly excites
  7. L6:1718 strongly excites
  8. L13:975 strongly excites
  9. L19:663 excites
  10. L8:393 excites
  11. L14:280 suppresses
  12. L21:716 excites
  13. L15:557 suppresses
  14. L7:111 excites
  15. L20:519 suppresses
  16. L11:810 suppresses
  17. L16:2520 excites
  18. L11:233 excites
  19. L16:715 excites
  20. L8:2755 excites
  21. L6:3016 suppresses
  22. L21:274 suppresses
  23. L6:199 excites
  24. L7:2164 excites period inside country abbreviations

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 -1.99 and -1.881 values between -1.88 and -1.761 values between -1.76 and -1.652 values between -1.65 and -1.541 values between -1.54 and -1.431 values between -1.43 and -1.322 values between -1.32 and -1.219 values between -1.21 and -1.103 values between -1.10 and -0.9911 values between -0.99 and -0.8815 values between -0.88 and -0.7628 values between -0.76 and -0.6534 values between -0.65 and -0.5461 values between -0.54 and -0.43120 values between -0.43 and -0.32195 values between -0.32 and -0.21408 values between -0.21 and -0.101067 values between -0.10 and 0.01660 values between 0.01 and 0.13329 values between 0.13 and 0.24391 values between 0.24 and 0.35263 values between 0.35 and 0.46148 values between 0.46 and 0.5778 values between 0.57 and 0.6831 values between 0.68 and 0.7915 values between 0.79 and 0.908 values between 0.90 and 1.013 values between 1.01 and 1.138 values between 1.13 and 1.243 values between 1.24 and 1.353 values between 1.35 and 1.462 values between 1.57 and 1.681 values between 1.68 and 1.792 values between 1.79 and 1.901 values between 2.01 and 2.131 values between 2.35 and 2.46offnegativepositiveoften

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. ものが
  2. two
  3. 时候
  4. 時候
  5. 것
  6. 口气
  7. 愿望
  8. ことが
  9. 것도
  10. combination

Suppresses

  1. lessly
  2. less
  3. ential
  4. ously
  5. ocop
  6. acea
  7. otent
  8. iless
  9. omorphic
  10. ABEL