Neuron Explorer

Qwen3-0.6B

line break splitting a 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 fires on a line break that cuts a sentence in the middle, where the words before it are unfinished and the next line will complete them. When it was turned up, the model's writing fell into repetitive loops, so the steering results tell us little about it.

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

    century will be presented.⏎Mining-industry; Underground-mining; Coal-mining; Mine-safety; Ventilation; Coal-dust; Dust-control; Control-technology⏎National Institute for Occupational Safety and Health, Pittsburgh Research Laboratory, Pittsburgh,

  2. #2 peak

    Cancer Center⏎Case Western Reserve University⏎DNA in its native form is composed of two molecules that have the⏎characteristic of being complementary such that one strand associates⏎with the other in a

  3. #3 peak

    ⏎Biology, chemistry, ecology, and physics are just a few of the sciences⏎involved in aquarium keeping. A classroom aquarium can be used to teach⏎students about specific topics such as anatomy or more complex topics such⏎as the food

  4. #4 peak

    half, and half again, to⏎make four smaller rectangles. How many different ways can you fold⏎This problem focuses on Dienes' Logiblocs. What is the same and

  5. #5 peak

    catalog of your choice, fill out an order form for merchandise as if you⏎intended to place an order. Share the completed form with your counselor and discuss⏎- With your parent's permission, locate at least five Web sites that are helpful

  6. #6 peak

    their community. Nowhere do the neo-nazi skinheads by themselves⏎constitute a mass-movement. The danger posed by these young racist thugs lies⏎in their taste for violence. Skinheads across the globe have committed numerous⏎mur

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 have been, I think, the first one in the case case and first case case's case, first case case's second case's case first case and first case's second case's first second case's case second second second second's

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

    Volcanic News Erutpion states it as being restless. http://volcansvanuatueruptionsgb.blogspot.fr/2014/03/08032014-nevado-de-col

  2. #2 peak

    and then its right child.⏎T / \ I S / \ D E<|endoftext|>Original URL: http://www.theregister.co.uk/2006/08/21/mars_geysers/⏎Martian pole freck

  3. #3 peak

    lights, laminar flow cabinets, and other biological safety cabinet components protect.⏎Read More @ https://cmibloginsight.blogspot.com/2023/05/biological-safety-cabinets-ensuring.html<|endoftext|>World

  4. #4 peak

    in business.⏎This week’s article by Jeremy Rosen on Rav Lichtenstein is available from http://jeremyrosen.blogspot.co.uk/2015/08/rav-lichtenstein-zl.html⏎Your comment will be posted

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. L10:77 suppresses
  2. L14:526 excites
  3. L14:6 suppresses
  4. L14:402 excites
  5. L14:732 excites
  6. L4:1097 suppresses
  7. L7:195 suppresses 'of' in set noun phrases
  8. L9:7 excites
  9. L14:2636 suppresses
  10. L9:190 excites
  11. L6:0 weakly excites
  12. L8:395 weakly suppresses
  13. L14:3 weakly excites
  14. L13:695 weakly excites
  15. L1:2 weakly suppresses
  16. L14:2553 weakly excites
  17. L8:230 weakly suppresses
  18. L10:486 weakly suppresses
  19. L9:61 weakly excites
  20. L14:2738 weakly excites
  21. L14:493 weakly excites
  22. L7:1310 weakly excites
  23. L6:1540 weakly excites
  24. L10:305 weakly excites

Feeds later layers

  1. L16:2520 suppresses
  2. L16:1258 suppresses
  3. L17:153 excites
  4. L17:151 suppresses
  5. L16:1374 suppresses
  6. L18:783 excites
  7. L20:1388 suppresses
  8. L25:1465 excites
  9. L17:553 suppresses
  10. L21:1003 suppresses
  11. L20:73 excites
  12. L17:359 suppresses
  13. L21:274 excites
  14. L18:1793 excites
  15. L20:1393 excites
  16. L17:2928 excites
  17. L21:2762 suppresses
  18. L24:1926 excites
  19. L20:1067 weakly suppresses
  20. L21:716 weakly suppresses
  21. L17:2953 weakly excites
  22. L17:1516 weakly suppresses
  23. L16:337 weakly suppresses
  24. L21:348 weakly suppresses mid-sentence line breaks

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 -4.41 and -4.131 values between -3.86 and -3.591 values between -3.59 and -3.311 values between -3.31 and -3.041 values between -3.04 and -2.762 values between -2.76 and -2.493 values between -2.49 and -2.225 values between -2.22 and -1.946 values between -1.94 and -1.6711 values between -1.67 and -1.3911 values between -1.39 and -1.1229 values between -1.12 and -0.8531 values between -0.85 and -0.57123 values between -0.57 and -0.30847 values between -0.30 and -0.031919 values between -0.03 and 0.25600 values between 0.25 and 0.52172 values between 0.52 and 0.8067 values between 0.80 and 1.0731 values between 1.07 and 1.3417 values between 1.34 and 1.624 values between 1.62 and 1.899 values between 1.89 and 2.178 values between 2.17 and 2.442 values between 2.44 and 2.711 values between 2.71 and 2.991 values between 2.99 and 3.262 values between 3.54 and 3.811 values between 3.81 and 4.081 values between 6.27 and 6.55offnegativepositiveoften

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. less
  2. osphere
  3. -up
  4. lessness
  5. enci
  6. 头
  7. eward
  8. amment
  9. ulation
  10. out

Suppresses

  1. theirs
  2. .”⏎⏎⏎⏎
  3. )**
  4. flate
  5. .\"
  6. ."⏎⏎⏎⏎
  7. ?"⏎⏎⏎⏎
  8. ]**
  9. ucas
  10. ** ⏎