Neuron Explorer

Qwen3-0.6B

line breaks mid-sentence

A neuron in layer 23 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 appears in the middle of a sentence, where the text is about to continue on the next line. Turning it up made the model's output fall into repeated, broken word fragments.

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

    99⏎African American Nurses have long history with the American Red -⏎Undeterred by initial rejection, Davis persisted and in l9l7 became the⏎first

  2. #2 peak

    of starting work, while 13% of graduates expected to obtain a management role within a year.2⏎Ambitious employees are attractive to the majority of⏎A survey by the British Chambers of Commerce highlights that firms believe that the majority of

  3. #3 peak

    avoid using words that⏎may have a frightening connotation such as "fire" and "burn."⏎Instead, in a straightforward manner, tell your child that the⏎deceased body, enclosed in a casket or container, is taken

  4. #4 peak

    6.⏎Among the ACWAs⏎innovations were the 40-hour work week,⏎unemployment insurance, and the creation of labor-owned housing and⏎banks (including the Amalgamated Bank of Chicago

  5. #5 peak

    projection.⏎15. Detail Drawing⏎Contains detail<|endoftext|>Rheumatic fever is an⏎inflammatory disease that may develop after an infection (strep throat or⏎scarlet fever) with group A Streptococcus

  6. #6 peak

    Shop #5. She led 16 seamstresses out of the plant in⏎protest because their piece rate was reduced. Hillman was among the⏎citys 40,000 garment workers who joined them. The

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: The effect of a perfect cross between a right and left side of a perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect perfect

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

    version 7 or higher has "version information" which has information of version itself.⏎next page : How to encode - Capacity. Encoding to data code words.⏎ next page<|endoftext|>Duck Stamps — or Migratory Bird Hunting and Conservation

  2. #2 peak

    still in use today The Compendious Book on Calculation by Completion and Balancing was written by Al-Khwārizmī (790-840) and the Arabic title of the book gave Europe the word “algebra

  3. #3 peak

    major site of the Nabataean civilisation, in the south of its zone of influence. Its integrity is remarkable and it is well cons

  4. #4 peak

    toes. Second, children can be distracted by far more things than physical teachers are used to. Add video calls, various methods of communication and chatting, or simply spending too much time trying to figure out the technology involved, and it can be pretty

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. L20:433 excites
  2. L21:173 excites
  3. L10:29 suppresses
  4. L22:193 excites
  5. L14:348 suppresses
  6. L21:219 excites
  7. L5:5 suppresses digit after a label like paper or type
  8. L7:8 excites
  9. L10:33 suppresses
  10. L11:31 suppresses
  11. L7:6 suppresses
  12. L4:52 weakly excites
  13. L15:646 weakly excites
  14. L6:101 weakly suppresses
  15. L8:21 weakly suppresses
  16. L20:959 weakly suppresses
  17. L3:20 weakly excites
  18. L16:558 weakly excites
  19. L9:7 weakly excites
  20. L21:395 weakly suppresses
  21. L15:286 weakly excites
  22. L8:32 weakly excites
  23. L20:2061 weakly excites
  24. L20:161 weakly excites

Feeds later layers

  1. L27:635 excites
  2. L27:71 suppresses
  3. L25:222 excites
  4. L26:224 excites
  5. L24:51 excites
  6. L26:17 excites
  7. L27:63 suppresses closing parenthesis in data lists
  8. L25:292 excites
  9. L27:323 excites
  10. L27:838 excites
  11. L27:954 excites
  12. L25:15 suppresses
  13. L26:103 excites blank space before a number
  14. L27:112 excites
  15. L26:98 excites
  16. L27:483 excites
  17. L26:396 excites
  18. L26:712 suppresses
  19. L25:112 excites
  20. L26:32 suppresses
  21. L25:814 excites
  22. L27:1373 excites
  23. L25:304 excites
  24. L27:238 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 -18.25 and -17.271 values between -15.32 and -14.354 values between -13.37 and -12.397 values between -12.39 and -11.427 values between -11.42 and -10.4413 values between -10.44 and -9.4615 values between -9.46 and -8.4933 values between -8.49 and -7.5154 values between -7.51 and -6.5462 values between -6.54 and -5.5695 values between -5.56 and -4.58139 values between -4.58 and -3.61204 values between -3.61 and -2.63229 values between -2.63 and -1.66316 values between -1.66 and -0.681678 values between -0.68 and 0.30462 values between 0.30 and 1.27187 values between 1.27 and 2.2595 values between 2.25 and 3.2387 values between 3.23 and 4.2064 values between 4.20 and 5.1841 values between 5.18 and 6.1533 values between 6.15 and 7.1317 values between 7.13 and 8.1118 values between 8.11 and 9.0810 values between 9.08 and 10.0619 values between 10.06 and 11.045 values between 11.04 and 12.014 values between 12.01 and 12.993 values between 12.99 and 13.962 values between 13.96 and 14.942 values between 15.92 and 16.891 values between 19.82 and 20.80offnegativepositiveoften

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. 's
  2. #
  3. *
  4. 1
  5. 2
  6. gil
  7. -
  8. 0
  9. 4
  10. BN

Suppresses

  1. .$
  2. herself
  3. ','$
  4. :%
  5. =",
  6. :
  7. -:
  8. cols
  9. domicile
  10. %.