Neuron Explorer

Qwen3-0.6B

comma or full stop ending a clause

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 commas that close a phrase in web boilerplate, like 'to continue reading this article, subscribe' or 'for other uses, see'. Turning it up made the model output nothing but stops and short words like 'So.', which fits the punctuation it promotes, though the text is broken.

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

    solar storm would still pose a risk for wandering astronauts, he adds, but there would be time to return to a shielded base.⏎To continue reading this article, subscribe to receive access to all of newscientist.com, including 20

  2. #2 peak

    about how to correct material in RePEc.⏎For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: (Michael McNulty)⏎If references are entirely missing, you can add

  3. #3 peak

    www.gpi.org.⏎Answers: C, A, B, B, C<|endoftext|>This article is about the Great American Novel as a concept. For other uses, see Great American Novel (disambiguation).⏎The "Great American Novel"

  4. #4 peak

    you think you should have access to this title, please contact your librarian.⏎To troubleshoot, please check our FAQs, and if you can't find the answer there, please contact us.<|endoftext|>Indigenous Peoples and Climate Change: from victims to

  5. #5 peak

    those involved with their care usually don’t ask about sleep problems and often pay little heed to the psychological issues underlying obesity,” study researcher G. Neil Thomas, PhD, of the department of public health, epidemiology and biostatistics at the

  6. #6 peak

    . For example, if the array is [1, 2, 3, 4, 5], it will print it in reverse order i.e. from the last

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: there is a new application for a new type of material in the city of Newbury-Wilton in Britain’s Northern Northern Highlands in the year 2022 BWC22B as part of a National New material development and

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

    of data, and our research aims to use statistical techniques and other methods such as artificial intelligence to interpret large data sets.⏎“While we have this mathematical and statistical knowledge, what’s missing is the framework that can capture all of this complex

  2. #2 peak

    microbes, while cancer researchers have used them to make important discoveries in cancer biology. Several new centres will be established during 2020.⏎“With our facilities, researchers can explore cells, which are a universe in themselves,” Eriksson says

  3. #3 peak

    into organic matter - for each of the sites from data on wood, leaf and fine-root production. The results proved surprising.⏎"When we got together to analyze these data, we expected to spend our time explaining the differences between sites," said Norby

  4. #4 peak

    ilage Damage⏎It can take much longer for cartilage to heal when it is damaged due to the fact that it does not have a blood supply. For some people, medications and exercise can help to reduce symptoms and improve the damage. In other

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. L20:637 strongly suppresses
  2. L20:756 strongly excites
  3. L18:2501 strongly excites
  4. L17:1299 strongly excites
  5. L17:1048 strongly suppresses
  6. L16:768 strongly excites
  7. L21:148 strongly excites
  8. L18:1179 strongly suppresses
  9. L20:82 suppresses
  10. L21:102 suppresses
  11. L17:449 excites
  12. L16:1826 excites
  13. L20:210 excites
  14. L17:735 excites
  15. L21:344 suppresses
  16. L20:373 suppresses
  17. L20:1304 excites
  18. L22:417 suppresses period closing a web blurb
  19. L20:1957 suppresses
  20. L15:2505 excites
  21. L15:2235 suppresses
  22. L15:1159 excites ampersand in citation author lists
  23. L19:2931 suppresses
  24. L8:174 excites

Feeds later layers

  1. L27:975 excites
  2. L27:366 excites
  3. L26:775 excites
  4. L27:319 excites
  5. L26:255 excites
  6. L26:421 excites
  7. L26:11 suppresses
  8. L26:373 suppresses
  9. L27:646 excites
  10. L26:449 suppresses
  11. L27:703 excites
  12. L26:62 suppresses
  13. L24:12 suppresses
  14. L26:481 excites
  15. L27:347 excites
  16. L26:243 suppresses
  17. L26:122 excites
  18. L27:891 suppresses
  19. L25:430 excites
  20. L26:648 suppresses
  21. L27:385 excites
  22. L26:864 excites
  23. L27:25 suppresses
  24. L24:736 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 -21.58 and -20.824 values between -20.07 and -19.313 values between -19.31 and -18.564 values between -18.56 and -17.805 values between -17.80 and -17.044 values between -17.04 and -16.2915 values between -16.29 and -15.539 values between -15.53 and -14.7813 values between -14.78 and -14.0221 values between -14.02 and -13.2624 values between -13.26 and -12.5143 values between -12.51 and -11.7575 values between -11.75 and -11.0074 values between -11.00 and -10.2497 values between -10.24 and -9.48118 values between -9.48 and -8.73171 values between -8.73 and -7.97170 values between -7.97 and -7.22193 values between -7.22 and -6.46198 values between -6.46 and -5.71212 values between -5.71 and -4.95237 values between -4.95 and -4.19234 values between -4.19 and -3.44234 values between -3.44 and -2.68197 values between -2.68 and -1.93200 values between -1.93 and -1.17201 values between -1.17 and -0.41395 values between -0.41 and 0.34447 values between 0.34 and 1.10272 values between 1.10 and 1.8532 values between 1.85 and 2.612 values between 2.61 and 3.371 values between 3.37 and 4.121 values between 4.88 and 5.631 values between 7.90 and 8.66offnegativepositiveoften

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. .
  2. !
  3. !:
  4. .;
  5. ;
  6. !.
  7. ?
  8. .?
  9. :
  10. .:

Suppresses

  1. ẞ
  2. _then
  3. illions
  4. ultip
  5. gebn
  6. 的人来说
  7. 我说
  8. ’e
  9. “Oh
  10. звуч