Neuron Explorer

Qwen3-0.6B

comma in a list or after a noun phrase

A neuron in layer 21 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 mostly fires on commas and small joining words that come after a noun phrase, where a list or a longer description is about to continue. It has no clear topic. Turning it up made the model produce unrelated sentences about wind and streets.

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

    Probably not. Hoarding disorder goes beyond stockpiling in an emergency. I am a psychiatrist at the University of Florida and the director of the Center for OCD, Anxiety and Related Disorders. I also recently authored a book on hoarding disorder.

  2. #2 peak

    curious readers."⏎—Andrew Piper, McGill University⏎"If there is one book to assign Humanities and Digital Humanities students that provides a wide range of knowledge about text technologies, it is this one. It introduces principal concepts along with ample historical examples, five

  3. #3 peak

    and analytical purposes. It is because the re-exported goods might require some additional information.⏎Having understood so much about reexport trade and what are re-exportation, it is<|endoftext|>10.2.2 Restarting TCP/IP⏎TCP/IP

  4. #4 peak

    . While there has been remarkable progress worldwide in giving millions of children, including girls, access to primary education, 75 million children never receive any

  5. #5 peak

    the majority of whom are either in pre-primary or primary education. With two of its programmes focusing on Life Skills and Literacy, a third on Early Childhood Development and a fourth on Nutrition, it was inevitable that the teams involved would be keen to

  6. #6 peak

    application using XCOPY will not copy the settings in the machine configuration file.⏎For more information about how the common language runtime uses the machine configuration file for assembly binding, see How the Runtime Locates Assemblies.⏎An application configuration file contains settings that

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: what if we have a way to show that a certain value, say, 3, is not 4? How would we do that? I can't tell you all of the details, but I can tell you that it is 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

    at the moment of Austen’s birth or there about, John Adams left his farm in Massachusetts for the Continental Congress in Philadelphia. Doesn’t sound particularly interesting, I know but consider this.⏎John Adams left his home in mid-December 1

  2. #2 peak

    , B-12, and B- 6. Milk has also been shown to improve weight and bone density in children, and reduce fractures. Also research has found that pregnant women who had diets rich in milk and dairy had babies with

  3. #3 peak

    full human experience’ requires the maximization of the incorporeal spirit, with minimal focus on the physical body, though whether this remains a practically viable pursuit, I can not agree.⏎Now how does this hypothesis compare to that of Homer? Having

  4. #4 peak

    monomers can be linked together in different ways, resulting in different types of plastic. The monomers used to make PET bottles are derived from petroleum, and there have been some concerns that chemicals from the plastic could leach into the contents of 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. L18:928 suppresses
  2. L20:1214 suppresses
  3. L20:2366 excites
  4. L20:80 suppresses
  5. L20:1595 suppresses
  6. L20:277 excites
  7. L19:1554 excites
  8. L19:91 excites
  9. L20:858 suppresses
  10. L20:264 excites
  11. L20:2998 suppresses
  12. L20:1003 suppresses
  13. L17:2277 suppresses
  14. L16:896 suppresses
  15. L8:892 suppresses
  16. L19:423 suppresses words that set up a following clause
  17. L19:2970 excites
  18. L19:351 excites
  19. L20:73 weakly suppresses
  20. L14:312 weakly suppresses
  21. L19:954 weakly suppresses
  22. L18:419 weakly excites
  23. L18:243 weakly excites
  24. L20:1043 weakly suppresses

Feeds later layers

  1. L26:194 weakly suppresses
  2. L22:225 weakly excites
  3. L22:241 weakly suppresses
  4. L26:87 weakly excites
  5. L22:61 weakly suppresses
  6. L27:556 weakly suppresses
  7. L27:326 weakly excites
  8. L27:1013 weakly excites
  9. L24:174 weakly excites
  10. L24:2065 weakly excites
  11. L27:468 weakly suppresses
  12. L22:2239 weakly excites
  13. L23:1719 weakly excites
  14. L23:1574 weakly excites
  15. L22:932 weakly excites
  16. L26:547 weakly excites
  17. L25:2059 weakly suppresses
  18. L23:2084 weakly excites
  19. L23:2591 weakly excites
  20. L22:727 weakly excites
  21. L25:1745 weakly excites
  22. L27:1017 weakly excites
  23. L23:1917 weakly excites
  24. L22:108 weakly 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 -17.72 and -16.972 values between -16.22 and -15.471 values between -14.72 and -13.971 values between -13.97 and -13.222 values between -13.22 and -12.474 values between -12.47 and -11.717 values between -11.71 and -10.969 values between -10.96 and -10.2110 values between -10.21 and -9.469 values between -9.46 and -8.7118 values between -8.71 and -7.9625 values between -7.96 and -7.2133 values between -7.21 and -6.4643 values between -6.46 and -5.7158 values between -5.71 and -4.9676 values between -4.96 and -4.21113 values between -4.21 and -3.46142 values between -3.46 and -2.71182 values between -2.71 and -1.96325 values between -1.96 and -1.21517 values between -1.21 and -0.461578 values between -0.46 and 0.30394 values between 0.30 and 1.05157 values between 1.05 and 1.8092 values between 1.80 and 2.5540 values between 2.55 and 3.3027 values between 3.30 and 4.0515 values between 4.05 and 4.8012 values between 4.80 and 5.556 values between 5.55 and 6.304 values between 6.30 and 7.051 values between 7.05 and 7.801 values between 7.80 and 8.552 values between 10.05 and 10.801 values between 11.55 and 12.31offnegativepositiveoften

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. olesterol
  2. _battery
  3. upper
  4. immune
  5. ขน
  6. Compact
  7. serotonin
  8. ucle
  9. encryption
  10. vertical

Suppresses

  1. 对我说
  2. 越好
  3. 这样说
  4. 涑
  5. 又好
  6. cripts
  7. hof
  8. 也不例外
  9. 的例子
  10. 沭