Neuron Explorer

Qwen3-0.6B

line breaks mid-sentence

A neuron in layer 15 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 reacts to a hard line break that cuts a sentence in the middle, right before the rest of the phrase continues on the next line. When it was turned up, the model's writing drifted into unrelated topics like maths and the field of objects, so the effect is hard to read.

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

    of Booch, OMT and OOSE by fusing them into a single,⏎common and widely usable modelling language. UML aims to be a standard modelling⏎language which can model concurrent and distributed systems.⏎UML is not an industry standard

  2. #2 peak

    ital a piece of ground contiguous to Pardon Churchyard, called the Spital Croft, which the good Bishop Stratford also consecrated. The two burial-grounds, afterwards united, were known as New Church Hawe.⏎Stow,

  3. #3 peak

    ⏎the word itself, the OneLook Reverse Dictionary can help. The site indexes⏎hundreds of online dictionaries, encyclopedias, and other reference⏎sites for words that have definitions similar to the words you search⏎for. Developed

  4. #4 peak

    a level of 300 pmol/L. Understanding the clinical⏎manifestations of vitamin B12 deficiency and how it is relevant to Australia’s refugee⏎population might assist to resolve some of the difficulties that refugees face in Australia.⏎Download

  5. #5 peak

    this cult until⏎today it has supported violence, starting with beatings with boots and small⏎weapons, evolving today into a terrorist like cult armed with deadly military⏎weapons.Just like gangs in the U.⏎S. skinheads also fight among

  6. #6 peak

    the results of the first stage. As of September 2008 the design teams⏎use 3D modeling software to generate CAD drawings from initial design⏎sketches. The designs are then prototyped using an in-house⏎

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 will not. I am not. I can't. I am not. I will not. I will not. I don. I can't. I can't. 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

    program does not try to expunge the odd social behaviors that often make life so difficult for them. Its unconventional aim is to teach students that it is O.K. to ''act autistic'' and also how to get by in a world where it

  2. #2 peak

    we here at Housman Institute are reflecting on gratitude and what it means for the development of emotional intelligence in young children. As Robert Emmons, Ph.D. the world’s leading scientific expert on gratitude. a professor of psychology and founding editor

  3. #3 peak

    or parasites. Toxins produced by bacteria<|endoftext|>Dit artikel staat op icthealth.nl. Link: https://ictandhealth.com/ucla-looks-into-rewiring-brain-in-order-to-help-with-depressions/news⏎U

  4. #4 peak

    agency is סוכנות, most recognizably in the Hebrew name of The Jewish Agency – הַסּוֹכְנוּת הַיְּהוּד�

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. L9:1515 excites line break splitting a phrase
  2. L4:212 excites
  3. L12:792 excites
  4. L14:518 suppresses
  5. L14:409 suppresses
  6. L6:3016 suppresses
  7. L8:252 excites 'of' after 'each' or 'one' or 'all'
  8. L5:2594 suppresses
  9. L12:1500 suppresses
  10. L8:1208 suppresses
  11. L14:2636 suppresses
  12. L13:988 weakly excites
  13. L7:2935 weakly suppresses
  14. L7:2164 weakly suppresses period inside country abbreviations
  15. L14:35 weakly excites
  16. L6:181 weakly excites periods in abbreviations and the 's in names
  17. L12:51 weakly suppresses
  18. L14:51 weakly excites colon that introduces a statement
  19. L13:272 weakly suppresses
  20. L9:2455 weakly excites
  21. L8:395 weakly suppresses
  22. L13:131 weakly suppresses 'and' joining parallel phrases
  23. L8:101 weakly excites unclear: commas and 'or' in lists
  24. L6:972 weakly excites "well" after "how"

Feeds later layers

  1. L17:390 strongly excites
  2. L16:715 strongly excites
  3. L20:526 excites
  4. L18:1321 excites
  5. L18:305 excites
  6. L21:68 suppresses
  7. L20:1197 excites 'this' pointing at the current document
  8. L21:323 excites
  9. L21:498 excites
  10. L19:1009 excites
  11. L19:207 suppresses
  12. L16:1578 excites
  13. L18:2629 excites
  14. L16:2042 excites
  15. L21:348 excites mid-sentence line breaks
  16. L20:1212 excites
  17. L19:1155 excites
  18. L20:318 excites
  19. L17:510 excites
  20. L19:135 suppresses
  21. L27:227 excites
  22. L18:1326 excites
  23. L21:553 excites
  24. L20:337 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 -3.67 and -3.481 values between -2.36 and -2.171 values between -2.17 and -1.984 values between -1.98 and -1.792 values between -1.23 and -1.044 values between -1.04 and -0.8649 values between -0.86 and -0.67129 values between -0.67 and -0.48355 values between -0.48 and -0.29844 values between -0.29 and -0.111673 values between -0.11 and 0.08650 values between 0.08 and 0.27139 values between 0.27 and 0.4615 values between 0.46 and 0.647 values between 0.64 and 0.831 values between 0.83 and 1.024 values between 1.02 and 1.216 values between 1.21 and 1.392 values between 1.39 and 1.582 values between 1.58 and 1.775 values between 1.77 and 1.962 values between 1.96 and 2.143 values between 2.14 and 2.331 values between 2.33 and 2.524 values between 2.52 and 2.711 values between 3.27 and 3.461 values between 3.46 and 3.642 values between 3.64 and 3.83offnegativepositiveoften

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. series
  2. series
  3. 字样
  4. 角度来看
  5. management
  6. footing
  7. 항
  8. direction
  9. 식
  10. 条

Suppresses

  1. 恶
  2. бав
  3. Compose
  4. (surface
  5. Peel
  6. adle
  7. 奏
  8. нные
  9. Resolve
  10. Stam