Neuron Explorer

Qwen3-0.6B

'for' introducing an example or reason

A neuron in layer 17 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

The neuron fires on the word 'for' when it starts a phrase like 'for example', 'for instance' or 'for several reasons'. It is quiet, and slightly negative, on 'before' in phrases like 'But before that'. Steering results did not show a clear theme.

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

    ⏎The Negotiator Magazine November 2014<|endoftext|>In most homes, regardless of parenting style, arguing with the kids is not encouraged. In fact, for most of us, it

  2. #2 peak

    lead too far afield; but it is clear from all that is known of the<|endoftext|>Obese & Fat German Shepherd⏎Like humans, dogs can become obese for several reasons, and the effect is just as severe in them. German Shepher

  3. #3 peak

    /or less comfortable expressing themselves via words. Art Therapy is also known to help people with social skills, specifically, those that are withdrawn or shy, or who, for one reason or another, have a difficult time functioning within social situations.⏎Why art

  4. #4 peak

    levels of exposure to other cultures and those with low levels of domestic alternatives. Small countries that have experienced struggles with outside countries through political, military and social events, for example, are more likely to carry those negative feelings to the marketplace when it comes

  5. #5 peak

    understanding of each other’s environmental problems and to explore opportunities for positive action, individually and collectively.”⏎Other nations were far more skeptical. France and the United Kingdom, for example, were<|endoftext|>Diseases & Conditions⏎If you trim your toenails

  6. #6 peak

    .S., we have a number of ethnic nations, most, if not all, being Amerindian in nature. But, citizens of the Navajo nation, for instance, have their own police force and legal system. To be a member of

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: that that, in doing so, that, that, that, that, that, that, that, that, that, that, that, that, and that, that, that, that, that, and then, that, that

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

    Thomas Edison Pitch Contest⏎The Thomas Edison Pitch Contest was created in 2010 as a competition to showcase and encourage invention, innovation and entrepreneurship among students before they get to college. Every year, students submit their great ideas to win prizes

  2. #2 peak

    new growth add further ornamental interest. Sun-loving Japanese tree lil<|endoftext|>At present, there are several political parties in India are providing social responsibilities towards society. But before the party, we do not have known or seen the previous Congress, the Indian

  3. #3 peak

    winds will be fierce. The sun will be of no use. Three of these winters will come, one after the other, with no summer in between. But before that, there will have

  4. #4 peak

    the bin, consider feeding them to animals.<|endoftext|>Student Vote Program Evaluation⏎To educate young Canadians about democracy and elections and to help them develop the habit of voting before they turn 18, Elections Canada has worked with Student Vote to provide students

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. L16:2336 excites
  2. L9:1890 excites
  3. L4:1035 suppresses 'for' after verbs like 'fall' or 'click'
  4. L3:381 weakly excites 'for' in research institute names
  5. L10:132 weakly excites
  6. L8:705 weakly excites
  7. L15:1838 weakly suppresses
  8. L15:906 weakly suppresses
  9. L6:2320 weakly suppresses
  10. L0:2365 weakly suppresses mid-word pieces like 'advent' and 'precaution'
  11. L6:617 weakly suppresses
  12. L5:697 weakly suppresses
  13. L5:1090 weakly excites
  14. L5:501 weakly excites
  15. L3:2256 weakly suppresses
  16. L5:2269 weakly suppresses
  17. L15:280 weakly excites
  18. L11:1868 weakly excites
  19. L0:1183 weakly excites
  20. L15:1974 weakly excites
  21. L5:148 weakly suppresses
  22. L5:1041 weakly excites
  23. L11:210 weakly excites
  24. L5:273 weakly suppresses

Feeds later layers

  1. L21:1722 weakly excites
  2. L18:1617 weakly suppresses
  3. L18:2673 weakly excites
  4. L24:1243 weakly suppresses
  5. L18:1389 weakly suppresses
  6. L18:1775 weakly suppresses
  7. L19:1914 weakly suppresses
  8. L21:2473 weakly excites
  9. L23:1858 weakly suppresses
  10. L26:2344 weakly suppresses
  11. L23:2862 weakly suppresses
  12. L24:2210 weakly suppresses
  13. L27:1075 weakly suppresses
  14. L21:1546 weakly suppresses
  15. L20:2745 weakly excites
  16. L19:1050 weakly excites
  17. L23:508 weakly excites
  18. L19:2727 weakly excites
  19. L24:1352 weakly suppresses
  20. L21:1533 weakly suppresses
  21. L19:2786 weakly excites
  22. L21:741 weakly suppresses
  23. L18:1391 weakly excites
  24. L18:2938 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.

1 values between -14.62 and -14.042 values between -4.77 and -4.192 values between -4.19 and -3.614 values between -3.61 and -3.044 values between -3.04 and -2.467 values between -2.46 and -1.8810 values between -1.88 and -1.3022 values between -1.30 and -0.72783 values between -0.72 and -0.142964 values between -0.14 and 0.4487 values between 0.44 and 1.029 values between 1.02 and 1.603 values between 1.60 and 2.184 values between 2.18 and 2.761 values between 3.34 and 3.911 values between 3.91 and 4.491 values between 4.49 and 5.071 values between 5.07 and 5.651 values between 5.65 and 6.231 values between 7.97 and 8.55offnegativepositiveoften

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. ally
  2. альным
  3. iado
  4. away
  5. plays
  6. ulate
  7. erten
  8. holes
  9. альные
  10. aside

Suppresses

  1. sworth
  2. .fc
  3. FRAME
  4. úc
  5. fov
  6. 驷
  7. 亲眼
  8. еств
  9. จบ
  10. imits