Neuron Explorer

Qwen3-0.6B

comma after a finished clause

A neuron in layer 3 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

It fires on a comma that comes after a complete statement and before an extra detail, such as 'remains somewhat unclear,' or 'Copper naturally exhibits antibacterial properties,'. Turning it up did not produce a clearly related effect in the sample outputs.

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

    types possess a medulla, as its presence varies among individuals and even within different hair strands of the same person. The function of the medulla remains somewhat unclear, although it is believed to contribute to the overall strength and flexibility of the hair.⏎

  2. #2 peak

    such as money lenders and payment couriers) to meet their financial needs—from receiving wages to saving money for fertilizer. However, these informal mechanisms tend to be insecure, expensive, and complicated to use. And they offer limited recourse when major problems arise

  3. #3 peak

    , 13-year-old Jared High suffered from physical injuries, depression, and possibly post-traumatic stress disorder. He committed suicide on Sept. 29, 1998.⏎Bulling isn’t a game of “child’s

  4. #4 peak

    tremendous strain on electrical systems, as seen during Texas’s deadly winter storm. Summer blackouts may become even more common, and deadlier, as temperatures soar. The energy required to save lives during these

  5. #5 peak

    that 12.4 million Californians live in census tracts with high “social vulnerability” to climate change.⏎This vulnerability can play out in various ways, says Heather Cooley, co-director of the institute’s water program and a

  6. #6 peak

    from boom to bust about 37,000 years ago. The bison population drop began well before large numbers of humans lived in the New World, and this timing suggests that climate and environmental change — not human hunting pressure — drove

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: this is a question. Let me know. So, the question is, what is the relationship between the numbers and symbols. Okay, so the question is, what is the relationship between the numbers and symbols. So, the answer is,

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

    Bleeding-heart Gallicolumba platenae and Black-hooded Coucal Centropus steerii, and the Endangered Mindoro Hornbill Penelopides mindorensis. The Haribon

  2. #2 peak

    might not want to read about this giant new rat species which was discovered deep in the jungle of Papua New Guinea.⏎It is called the “Bosavi Woolly Rat” and was discovered during the filming of Lost Land of the Volcano,

  3. #3 peak

    , and by Nebuchadnezzar (586–573 BC) for thirteen

  4. #4 peak

    idemic” proportions. Studies show that the flu costs businesses approximately $7 billion per year due to the nearly 111 million work days lost. Norovirus cases are also on the rise this season, and the CDC reports that food

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. L2:66 excites
  2. L1:86 weakly excites
  3. L2:284 weakly suppresses
  4. L1:1377 weakly suppresses placeholder example email domain
  5. L1:1357 weakly excites
  6. L1:371 weakly excites
  7. L2:871 weakly suppresses
  8. L2:126 weakly excites
  9. L1:1440 weakly suppresses
  10. L0:146 weakly suppresses
  11. L1:1499 weakly excites
  12. L0:2921 weakly suppresses
  13. L0:2488 weakly suppresses
  14. L1:2413 weakly excites
  15. L2:44 weakly excites
  16. L0:1218 weakly excites
  17. L0:63 weakly suppresses
  18. L1:1869 weakly excites
  19. L1:648 weakly suppresses
  20. L1:2556 weakly excites
  21. L2:771 weakly excites
  22. L1:1530 weakly suppresses
  23. L0:2830 weakly suppresses
  24. L2:2482 weakly suppresses

Feeds later layers

  1. L6:904 excites
  2. L22:1283 weakly suppresses
  3. L10:1698 weakly suppresses
  4. L25:23 weakly suppresses
  5. L6:2992 weakly suppresses
  6. L11:769 weakly suppresses
  7. L6:2724 weakly excites
  8. L7:1160 weakly excites
  9. L11:2976 weakly excites
  10. L24:343 weakly excites
  11. L14:1248 weakly excites
  12. L5:2910 weakly suppresses
  13. L15:1775 weakly suppresses
  14. L25:887 weakly excites
  15. L20:1417 weakly excites
  16. L19:1608 weakly excites
  17. L19:1310 weakly excites
  18. L18:11 weakly suppresses
  19. L14:829 weakly excites
  20. L4:5 weakly excites
  21. L5:2942 weakly excites
  22. L25:968 weakly suppresses
  23. L25:655 weakly excites
  24. L5:3041 weakly 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.

2 values between -1.72 and -1.541 values between -1.18 and -1.002 values between -1.00 and -0.822 values between -0.82 and -0.6515 values between -0.65 and -0.47251 values between -0.47 and -0.291127 values between -0.29 and -0.111857 values between -0.11 and 0.07495 values between 0.07 and 0.2539 values between 0.25 and 0.4310 values between 0.43 and 0.6112 values between 0.61 and 0.7910 values between 0.79 and 0.9712 values between 0.97 and 1.155 values between 1.15 and 1.334 values between 1.33 and 1.512 values between 1.51 and 1.686 values between 1.68 and 1.866 values between 1.86 and 2.043 values between 2.04 and 2.223 values between 2.22 and 2.404 values between 2.40 and 2.585 values between 2.58 and 2.766 values between 2.76 and 2.942 values between 2.94 and 3.125 values between 3.12 and 3.302 values between 3.30 and 3.484 values between 3.48 and 3.652 values between 3.65 and 3.833 values between 3.83 and 4.014 values between 4.01 and 4.193 values between 4.37 and 4.552 values between 4.55 and 4.731 values between 4.91 and 5.091 values between 5.27 and 5.45offnegativepositiveoften

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. ident
  2. reconstruct
  3. redraw
  4. correct
  5. oding
  6. xaa
  7. integrate
  8. integrates
  9. ी
  10. ++]

Suppresses

  1. ()->
  2. 了下来
  3. ulary
  4. 了一系列
  5. empor
  6. ẢN
  7. laws
  8. ailable
  9. eced
  10. 实践活动