Neuron Explorer

Qwen3-0.6B

comma closing an intro clause

A neuron in layer 22 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 comma that ends an opening phrase like 'If the fan is not working,' or 'To understand this,' right before the main sentence begins. Turned up, the model got stuck repeating filler words like 'there' and 'the'.

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

    hazards, requirements for the right PPE, and tips and tricks.⏎Common Electrical Hazards⏎To discuss common electrical hazards that require the use of shockproof gloves, it’s important to first consider the nature of exposure

  2. #2 peak

    found that mast cells – immune cells involved in allergic responses – play a key role,” Lenz said.⏎If human development mirrors what was seen in this animal study, it’s possible that relatively minor influences – such as an allergic reaction, injury or

  3. #3 peak

    1 target of 95 g CO2/km is ambitious, compared with the current average. If this average remained the same past 2021, with an estimated violation fine of €

  4. #4 peak

    5 wagons in service, from a fleet of about 400 wagons.⏎With the start of railway rehabilitation in the 1950's, two underframes from scrapped wagons were utilised to build ballast pl

  5. #5 peak

    respectively. Confusingly some species — zebra, giraffe, gazelle and buffalo — have remained relatively stable. To understand such contrasting fortunes, scientists from the Dartmouth and Utah universities in the US are working to determine whether

  6. #6 peak

    evaporator fan motor. On refrigerators with only one evaporator, the evaporator is located in the freezer compartment. If the evaporator fan is not working, it will not circ

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: When and should to whom should to whom to whom should to whom and who should to whom and who should to whom and who should to whom and who should to whom and who should to whom and who should to whom and who should to whom

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

    20|⏎|Order Information:||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce

  2. #2 peak

    consecutive. How many times like this are there between midnight⏎and 7 a.m.?⏎These practical challenges are all about making a 'tray' and covering it with paper.⏎My cousin was 24 years old on Friday April

  3. #3 peak

    30% of behavioral intention, social norms are rarely even significant and always account for a very small percentage towards the variance of exercise intentions. ...⏎Not exactly what you need?<|endoftext|>Member of the Turkish National Allergy and Clinical Immunology Association

  4. #4 peak

    others would appreciate reading this, please share it through the social media buttons.⏎When Mary arrived and saw Jesus, she fell at his feet and said, “Lord, if only you had been here, my brother would not have died.”⏎“Where

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. L20:122 strongly excites
  2. L20:2430 strongly excites
  3. L17:2397 strongly excites
  4. L20:660 strongly excites
  5. L19:1972 strongly suppresses
  6. L21:191 strongly suppresses
  7. L17:1845 strongly excites
  8. L20:982 strongly excites
  9. L20:850 strongly excites
  10. L15:3010 excites
  11. L20:1959 suppresses
  12. L17:2162 suppresses
  13. L17:1426 excites
  14. L20:1832 excites 'as' introducing a reason
  15. L20:974 excites
  16. L19:2058 excites
  17. L17:3021 excites
  18. L20:276 suppresses
  19. L21:424 excites
  20. L18:776 suppresses
  21. L16:1365 excites
  22. L6:2843 suppresses
  23. L21:2045 excites unclear: weak noun and pronoun activity
  24. L11:949 excites

Feeds later layers

  1. L26:236 excites
  2. L23:298 excites
  3. L24:49 suppresses
  4. L26:22 suppresses the 'res' fragment inside words
  5. L27:717 excites
  6. L25:381 suppresses
  7. L27:358 excites
  8. L27:130 suppresses
  9. L25:247 excites
  10. L25:234 excites
  11. L24:216 excites
  12. L27:86 suppresses
  13. L27:121 suppresses
  14. L24:398 excites
  15. L25:328 excites
  16. L24:468 suppresses
  17. L25:438 excites
  18. L27:1095 excites
  19. L27:286 excites
  20. L25:538 excites
  21. L25:346 suppresses
  22. L25:1994 excites
  23. L26:844 suppresses
  24. L27:108 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 -11.21 and -9.841 values between -5.73 and -4.361 values between -4.36 and -3.0062 values between -3.00 and -1.632468 values between -1.63 and -0.26966 values between -0.26 and 1.11141 values between 1.11 and 2.4858 values between 2.48 and 3.8550 values between 3.85 and 5.2226 values between 5.22 and 6.5925 values between 6.59 and 7.9616 values between 7.96 and 9.339 values between 9.33 and 10.7010 values between 10.70 and 12.0712 values between 12.07 and 13.449 values between 13.44 and 14.818 values between 14.81 and 16.184 values between 16.18 and 17.557 values between 17.55 and 18.913 values between 18.91 and 20.285 values between 20.28 and 21.655 values between 21.65 and 23.022 values between 23.02 and 24.395 values between 24.39 and 25.766 values between 25.76 and 27.132 values between 28.50 and 29.871 values between 29.87 and 31.241 values between 31.24 and 32.611 values between 33.98 and 35.352 values between 35.35 and 36.721 values between 42.19 and 43.56offnegativepositiveoften

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. there
  2. we
  3. ,
  4. ,
  5. 我们
  6. it
  7. there
  8. ،
  9. the
  10. theres

Suppresses

  1. 面孔
  2. edly
  3. 的母亲
  4. 톱
  5. 的脸
  6. merce
  7. 何况
  8. 漱
  9. 在路上
  10. imax