Neuron Explorer

Qwen3-0.6B

comma joining a full clause

A neuron in layer 18 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 a comma that follows a complete statement and sets up an extra explanation, such as 'a decision based on a feature, and it leads to...'. It was quiet or negative on words that end a phrase right before a bracket or an aside, like 'F' or 'implies'. Turning it up or down gave loose, repetitive text with no clear 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

    . The root node represents the top of the tree, and it contains all the instances or data points. Each branch represents a decision based on a feature or attribute, and it leads to a child node. The child node contains the instances that were

  2. #2 peak

    developed a tiny, $0.20 temperature sensor that didn’t require a battery or constant source of wired electricity. 3D printing isn't new either, and already, researchers can print out human organs using biological "ink".⏎The UW

  3. #3 peak

    as developers navigate through various programming languages and frameworks to create powerful and intuitive applications.⏎The Significance of Coding in App Development⏎Coding is the backbone of app development, allowing developers

  4. #4 peak

    at the top of your screen near the toolbars. Then right-click and select AutoCAD. A long list should show up, displaying various toolbars containing different commands. The most popular toolbars that are used

  5. #5 peak

    produce economic benefits through the jobs they provide and the products they produce. There is no way I can catalog all those benefits. No single statistic can do them justice, but the following may give some idea of their scale: data on the amount of

  6. #6 peak

    color to highlight or downplay certain features of your outfit.⏎Pattern is another key element of fashion. Patterns can range from bold and graphic to subtle and understated,

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 I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I

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

    if they’re given time to acclimate to the warmer weather, that temperature raises slightly, to 37.6° C (99.7° F). You might recognize that

  2. #2 peak

    b. (of two or more attributes of an object) unable to belong to the object simultaneously; inconsistent.⏎5. (of positions, functions, ranks, etc.) unable to be held simultaneously by one person.⏎6. Med.of or pertaining

  3. #3 peak

    would ultimately be approved. In other words, as of the date of this writing, it is still illegal to kill goliath grouper.⏎As its name implies, the Goliath Grouper

  4. #4 peak

    are also many approximate conservation laws, which apply to such quantities as mass, parity, lepton number, baryon number, strangeness, hypercharge, etc. These quantities are conserved in certain classes of physics processes, but not in

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. L14:417 excites
  2. L8:27 excites
  3. L17:1772 excites
  4. L14:1142 suppresses
  5. L17:845 suppresses comma after an introductory phrase
  6. L7:8 suppresses
  7. L6:228 suppresses comma after a filler word
  8. L16:558 suppresses
  9. L6:429 excites
  10. L16:180 suppresses
  11. L8:187 suppresses
  12. L17:432 excites
  13. L14:177 suppresses
  14. L17:1548 suppresses
  15. L16:1545 suppresses
  16. L17:2706 excites
  17. L8:32 suppresses
  18. L16:509 suppresses
  19. L14:830 excites
  20. L6:296 excites
  21. L14:220 suppresses
  22. L10:182 suppresses
  23. L7:701 excites
  24. L17:347 suppresses full stops ending informative sentences

Feeds later layers

  1. L19:384 suppresses
  2. L19:204 excites
  3. L19:909 suppresses
  4. L19:278 excites
  5. L20:2 suppresses
  6. L20:317 excites
  7. L20:382 suppresses
  8. L19:2683 suppresses
  9. L19:94 excites
  10. L20:95 suppresses
  11. L19:272 excites
  12. L20:271 excites
  13. L19:330 suppresses
  14. L20:660 suppresses
  15. L19:695 excites
  16. L20:245 suppresses
  17. L21:99 weakly excites
  18. L19:1072 weakly suppresses
  19. L19:270 weakly suppresses
  20. L19:719 weakly suppresses
  21. L23:335 weakly suppresses
  22. L20:252 weakly suppresses
  23. L21:639 weakly suppresses
  24. L19:36 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 -7.93 and -7.514 values between -7.09 and -6.663 values between -6.24 and -5.822 values between -5.82 and -5.395 values between -5.39 and -4.978 values between -4.97 and -4.547 values between -4.54 and -4.126 values between -4.12 and -3.7017 values between -3.70 and -3.2738 values between -3.27 and -2.8530 values between -2.85 and -2.4349 values between -2.43 and -2.0068 values between -2.00 and -1.58119 values between -1.58 and -1.15217 values between -1.15 and -0.73411 values between -0.73 and -0.312101 values between -0.31 and 0.12709 values between 0.12 and 0.5442 values between 0.54 and 0.9721 values between 0.97 and 1.396 values between 1.39 and 1.819 values between 1.81 and 2.249 values between 2.24 and 2.665 values between 2.66 and 3.083 values between 3.08 and 3.511 values between 3.51 and 3.933 values between 3.93 and 4.365 values between 4.36 and 4.784 values between 4.78 and 5.202 values between 5.20 and 5.632 values between 6.05 and 6.471 values between 8.59 and 9.02offnegativepositiveoften

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. t
  2. acman
  3. f
  4. addy
  5. wares
  6. ll
  7. communicator
  8. 为自己
  9. ling
  10. 색

Suppresses

  1. '): ⏎
  2. "): ⏎
  3. ��态
  4. ')['
  5. %");⏎
  6. '):⏎
  7. )::
  8. '), ⏎
  9. .'); ⏎
  10. '):