Neuron Explorer

Qwen3-0.6B

"a" before a descriptive noun phrase

A neuron in layer 17 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 word "a" in phrases like "develop a critical", "living in a rural" or "in a few days", where a description usually comes next. Turning it up did not produce a clear theme in the model's writing.

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

    - A great review for testing⏎- Answers on back of cards for easy self checking⏎Figurative Language Practice Cards provide an easy way to help students develop a critical reading comprehension skill. Each card consists of a leveled passage that contains a form

  2. #2 peak

    burg (30.991).⏎The majority of West Virginia's counties are designated as rural, with a small majority of the state's population living in a rural area. In 2010, the state had a <|endoftext|>

  3. #3 peak

    been added throughout this edition and rationales for the answers continue to be provided giving the students a better understanding of principles related to drug dosages. This fifth edition addresses the increasing

  4. #4 peak

    I am proud of everyone’s participation in our inaugural year of blogging. Please take a moment and check out some of our work.⏎The links below take you on a journey with my students. Each has a unique voice and experiences to share with you

  5. #5 peak

    .⏎Extinctions in fishes are mostly caused by loss of habitat and the introduction of nonindigenous species. In North America, there are more freshwater fish species in a typical drainage to the east of the Great Continental Divide than to the west, where

  6. #6 peak

    remember to take it.⏎To clear up your infection completely, use Ampicillin for the full course of treatment. Keep using it even if you feel better in a few days.⏎Ask your health care provider any questions you may have about how to

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: “If you, as in example, (for example, for example, for example, for example, for example, for example, for example, for example, for example, for example, for example, for example, for example,

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

    or longer. Late-onset asthma produces heavy inflammation of bronchial tubes and is the dominant cause of chronic asthma.⏎Differences in treatment⏎Discovering these two types of asthma reactions has caused doctors to modify their treatment plans. Drug companies as

  2. #2 peak

    in Fuhlsbüttel had already been in place for over half a century, when the Nazis⏎came to power in Germany⏎and began using it

  3. #3 peak

    , all of it connected to 128 gigabytes of shared memory. But the M1 Ultra is in fact a Frankenstein’s monster, consisting of two identical M1 Max chips bolted together using a silicon interface that serves as a

  4. #4 peak

    th cent. B.C., when it was first used, calligraphy has always been considered equal, or even superior, to painting. Chinese calligraphy began with a simplified seal script, known as "chancery script," in which the width

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. L4:137 excites
  2. L10:365 suppresses
  3. L16:341 excites
  4. L9:59 suppresses "a" meaning per (a day, a year)
  5. L3:2559 suppresses
  6. L8:7 excites
  7. L0:3035 suppresses
  8. L8:4 weakly excites
  9. L3:50 weakly excites
  10. L5:125 weakly excites "of" after all three, both, ten
  11. L16:880 weakly suppresses
  12. L6:7 weakly suppresses
  13. L5:169 weakly excites
  14. L5:110 weakly suppresses
  15. L11:2 weakly suppresses unclear: blank before a year or number
  16. L5:1127 weakly suppresses
  17. L15:114 weakly excites
  18. L6:12 weakly excites
  19. L11:2859 weakly suppresses
  20. L8:10 weakly excites
  21. L1:331 weakly suppresses
  22. L16:507 weakly suppresses
  23. L8:253 weakly excites
  24. L11:282 weakly excites

Feeds later layers

  1. L21:2472 suppresses
  2. L23:705 suppresses
  3. L18:973 excites
  4. L19:1375 suppresses
  5. L20:3051 suppresses
  6. L23:883 weakly suppresses
  7. L20:1736 weakly excites
  8. L26:2677 weakly suppresses
  9. L21:1318 weakly suppresses
  10. L26:2420 weakly suppresses
  11. L18:640 weakly excites
  12. L23:2176 weakly excites
  13. L25:1316 weakly suppresses
  14. L26:2601 weakly excites
  15. L19:1331 weakly suppresses
  16. L22:2883 weakly suppresses
  17. L27:2627 weakly excites
  18. L21:2486 weakly suppresses
  19. L22:2531 weakly suppresses
  20. L26:371 weakly suppresses
  21. L25:3034 weakly suppresses
  22. L18:285 weakly excites
  23. L23:712 weakly suppresses
  24. L24:815 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.

2 values between -1.46 and -1.282 values between -1.28 and -1.102 values between -1.10 and -0.923 values between -0.92 and -0.733 values between -0.73 and -0.5519 values between -0.55 and -0.37153 values between -0.37 and -0.191468 values between -0.19 and -0.012078 values between -0.01 and 0.17113 values between 0.17 and 0.3518 values between 0.35 and 0.539 values between 0.53 and 0.723 values between 0.72 and 0.903 values between 0.90 and 1.083 values between 1.08 and 1.263 values between 1.26 and 1.443 values between 1.44 and 1.621 values between 1.62 and 1.802 values between 1.80 and 1.983 values between 1.98 and 2.174 values between 2.17 and 2.351 values between 2.53 and 2.711 values between 2.71 and 2.892 values between 2.89 and 3.072 values between 3.07 and 3.251 values between 3.25 and 3.432 values between 3.43 and 3.621 values between 3.80 and 3.981 values between 4.16 and 4.341 values between 5.06 and 5.251 values between 5.61 and 5.79offnegativepositiveoften

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. steller
  2. ees
  3. alist
  4. anan
  5. anne
  6. esk
  7. -resistant
  8. edith
  9. 有一种
  10. onica

Suppresses

  1. sembles
  2. ido
  3. 寒假
  4. 捧
  5. änd
  6. world
  7. Cheer
  8. {};⏎
  9. {⏎⏎
  10. .unpack