Neuron Explorer

Qwen3-0.6B

the letter h starting a word fragment

A neuron in layer 27 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 separate 'h' that starts a word, like the 'h' in 'haram', 'hoods' or 'hikers' when the word is split. Turning it up made the model output long runs of quotation marks and line breaks.

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

    - Vegetable ingredients but those which cause intoxication⏎- Non-alcoholic food⏎For halal food guidelines, it is also important to know what foods are considered haram. So, you can avoid them. Here are all that you can consider

  2. #2 peak

    (not in the fume hood). Furthermore, it is best to keep the work space neat and clean in all laboratory operations, particularly those involving the use of hoods, so that any procedure or experiment can be undertaken

  3. #3 peak

    ’s one she and life is pronominally simple.⏎But things get tricky when she and she or he and he or singular, non-binary they and they rent huts:⏎Sandra and Sally rented a hut on Velika planina. She

  4. #4 peak

    and several others.They get support from so called punks and football (soccer) hooligans known as⏎"hools" to create riots and violence. From the

  5. #5 peak

    mountain side or strolling down a wooded path, hiking can feel like a good workout, but how many calories does is really burn?⏎FitDay explains that some hikers can burn up to 500 calories an hour while

  6. #6 peak

    , and check it every time the hood is used.⏎(For further information, see section 8.C.5 on testing and verification.)⏎Laboratory fume hoods and adjacent work areas should be kept clean and free of debris at all times

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: Wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear wear

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

    |ISBN (Print)||9 781742 589985|⏎|Publication status||Published - 2015|<|endoftext|>writing essay shiz⏎Word count: 350.⏎Need to

  2. #2 peak

    ||Web: http://www.tandfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce:v:42:y:2

  3. #3 peak

    LEE|⏎|Place of Publication||Abingdon, Oxon|⏎|Publication status||E-pub ahead of print - Aug 2020|<|endoftext|>10. Rebinding an Old Book Tutorial⏎Got an old book that

  4. #4 peak

    is common or similar in ideas or experiences. (151)⏎"It leads to light because it stems from the need to see: that which is not obvious...is to be transferred," (154

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. L24:2478 strongly excites
  2. L20:110 excites
  3. L22:1533 suppresses
  4. L25:36 suppresses
  5. L19:118 excites
  6. L18:541 suppresses
  7. L16:37 excites
  8. L4:489 excites
  9. L21:2374 excites
  10. L23:474 excites
  11. L20:2984 suppresses
  12. L20:111 suppresses
  13. L16:158 excites
  14. L0:304 excites
  15. L26:153 suppresses
  16. L3:516 excites
  17. L16:139 suppresses
  18. L26:534 excites
  19. L16:78 excites
  20. L16:95 suppresses
  21. L16:90 suppresses
  22. L19:189 excites
  23. L21:571 excites
  24. L16:0 excites

Feeds later layers

None: this is the last layer.

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 -16.391 values between -16.39 and -11.213 values between -11.21 and -6.0241 values between -6.02 and -0.833861 values between -0.83 and 4.351 values between 180.69 and 185.88offnegativepositiveoften

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. ız
  2. orses
  3. ög
  4. оро
  5. Wear
  6. inges
  7. ourg
  8. wnd
  9. ierarchy
  10. 可想而知