Neuron Explorer

Qwen3-0.6B

unclear: mixed possessives and fragments

A neuron in layer 19 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a grab bag of words, mostly ones that come right before a noun, such as 'its', 'their' or 'your' in business and web text. It is hard to say what ties them together. When it was turned up, the model's text drifted into rambling, repetitive phrases with no clear link to the neuron.

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

    is meant by "hashing function"? How is it used in hash data clusters and hash joins?⏎By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.⏎A hashing function

  2. #2 peak

    you know a student who could benefit from or may need assistive technology support or services?Your building's core team can help! Every district in Grant Wood AEA also has AT team members to support building core teams in designing accommodations that best match

  3. #3 peak

    these conditions for America's working men and women by setting and enforcing standards, and providing training, education and assistance. For more information, visit www.osha.gov.<|endoftext|>End of the year math review for 2nd grade can be

  4. #4 peak

    tree will<|endoftext|>Safety Data Sheet Information⏎Employers must ensure that the Safety Data Sheet (SDSs) are readily accessible to employees for all hazardous chemicals in their workplace. This may be done in many ways. For example, employers may keep

  5. #5 peak

    every day. Leaning against the ears, so there is almost no harm to the human body.<|endoftext|>DUARTE, Calif.--(BUSINESS WIRE)--The American Association for Cancer Research (AACR) released its inaugural Cancer Dispar

  6. #6 peak

    as well,” said Kevin Krewell, principal analyst, Tirias Research. “AMD made an audacious promise fifteen months ago to improve the energy efficiency of their processors by 25 times by 2020. A projected

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 United States of America have the right to the United States Air Service for the United States Air Force, the United States Air Agency, the United States Air Service for the United States Air Department, and the United States Air Force Service Association,

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

    37's.⏎After the fall of Poland, the German military incorporated 20 captured 7TPs into Tank Battalion 203 of the 1st Panzer Division, as the PzKpfw 73

  2. #2 peak

    findings. The project, known as Bolshoi, is led by Anatoly Klypin of New Mexico State University and Joel Primack of the University of California, Santa Cruz

  3. #3 peak

    Marxism/Leninism. Although Albania operated under the facade of constitutional rule, the communist party, led by Enver Hoxha, who was also president of Albania, actually controlled all aspects of the political, social, and economic systems.

  4. #4 peak

    .⏎The greatest president that has ever served the United States is George Washington. He was not only the first president of the United States but also the commander in chief of the Continental Army during the American Revolutionary War. He was victorious over the British and

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. L18:1227 excites
  2. L18:1510 suppresses
  3. L16:1830 excites
  4. L18:2529 excites
  5. L18:1931 suppresses
  6. L16:2810 suppresses
  7. L10:2558 suppresses
  8. L18:1667 suppresses
  9. L17:2747 excites
  10. L17:2086 suppresses
  11. L8:2585 suppresses
  12. L18:2012 excites
  13. L9:763 excites
  14. L10:1205 suppresses
  15. L12:2889 suppresses
  16. L18:2375 excites
  17. L16:2852 suppresses 'in' after a noun or participle
  18. L17:1522 suppresses
  19. L18:2072 excites
  20. L18:1507 excites
  21. L17:3041 excites
  22. L10:1318 excites
  23. L11:1994 excites
  24. L17:978 excites

Feeds later layers

  1. L22:2837 suppresses
  2. L21:2846 suppresses 'at' after a job title, before an institution
  3. L22:2134 suppresses
  4. L22:712 suppresses
  5. L20:2605 excites
  6. L27:886 suppresses
  7. L20:2778 suppresses
  8. L24:1655 suppresses
  9. L25:3048 suppresses
  10. L26:995 suppresses
  11. L20:2381 suppresses
  12. L26:1299 suppresses
  13. L22:2787 suppresses
  14. L24:2720 suppresses
  15. L23:2697 suppresses
  16. L20:1247 suppresses
  17. L21:372 suppresses
  18. L21:2944 suppresses
  19. L24:2460 suppresses
  20. L20:893 suppresses
  21. L24:1208 suppresses
  22. L24:1049 suppresses
  23. L21:1244 excites
  24. L23:1257 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 -20.00 and -19.291 values between -10.74 and -10.022 values between -8.60 and -7.882 values between -7.88 and -7.172 values between -7.17 and -6.466 values between -6.46 and -5.753 values between -5.75 and -5.039 values between -5.03 and -4.3211 values between -4.32 and -3.6120 values between -3.61 and -2.9029 values between -2.90 and -2.1837 values between -2.18 and -1.4785 values between -1.47 and -0.76713 values between -0.76 and -0.052744 values between -0.05 and 0.67199 values between 0.67 and 1.3816 values between 1.38 and 2.099 values between 2.09 and 2.819 values between 2.81 and 3.524 values between 3.52 and 4.233 values between 4.23 and 4.941 values between 5.66 and 6.371 values between 6.37 and 7.081 values between 7.80 and 8.51offnegativepositiveoften

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. 患者
  2. 人参
  3. ancellor
  4. 一杯
  5. eBook
  6. başv
  7. ToJson
  8. resumes
  9. 瓢
  10. 地图

Suppresses

  1. company
  2. 公司的
  3. 公司
  4. company
  5. companies
  6. Company
  7. 公司在
  8. 团队
  9. Company
  10. 团