Neuron Explorer

Qwen3-0.6B

"of" or "on" after part, face or footprint

A neuron in layer 13 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 small words like "of" or "on" that come after nouns such as "part", "face", "footprint" or "set foot" and before a place, as in "set foot on the moon". Turned up, the output did not show a clear matching 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

    away. The light we see today, comes from a time 10 million years after the dinosaurs were extinct. It was long before any human walked the face of the Earth. And for all we know, this star has gone supernova many

  2. #2 peak

    Rights Act” of 1964. The whole world watched in awe and amazement as Neil Armstrong became the first human being to ever set foot on the moon. The

  3. #3 peak

    this is a great exercise. Split the class up into groups and have each group color a large panorama of a landscape. Place different landscapes and habitats in different parts of the room. Provide pictures to the students so that they have a general idea of

  4. #4 peak

    ’s also a lot of history. NASA is worried that without some guidance and agreement irreplaceable relics of the Space Age, such as Neil Armstrong’s first footprint on the Moon could be lost and sites needlessly disturbed. More than that, many

  5. #5 peak

    tool for building students' critical thinking skills. It can be used in class, assigned for homework, or offered as a challenge or reward activity. The best part of all: as students learn to complete the puzzles they will be building confidence in their

  6. #6 peak

    with the bond between man and woman.⏎Why is it so hard to get possessives and plurals right? Wait, no, that’s not my favorite part of the blog. And it’s not Pocahontas, or Confuc

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: in the direction of the functions in the time in the time in the numbers in the numbers in the case in the elements in the case in the elements in the elements in the elements in the elements in the elements in the elements in the elements

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

    ⏎other States. It was not until the 90s when the colonies were promoted to States and the Commonwealth was founded.⏎State corporativism is the norm in Portugal, and imports and exports are rare as the government's aim is still

  2. #2 peak

    USA status by meeting four core standards of sound urban forestry management: maintain a tree board or department, having a community tree ordinance, spending at least $2 per capita on urban forestry and celebrating Arbor Day.⏎Alison Oelschl

  3. #3 peak

    what distinguished Tupac's music⏎You're talking about an artist who came from the people and decided that his work was going to reflect the conditions that were going on in America during his lifetime — his short 25 years on this planet.

  4. #4 peak

    ripe old age?<|endoftext|>Sixteen years ago, Patricia L. Gerbarg, a Kingston, N.Y., psychiatrist, contracted Lyme disease. But as sometimes happens, her blood tests were negative, so the condition was not diagnosed for five years

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. L12:2077 excites
  2. L12:2416 excites
  3. L12:850 excites
  4. L11:2022 excites
  5. L11:1316 weakly excites
  6. L12:2833 weakly excites
  7. L8:905 weakly excites
  8. L10:2020 weakly excites
  9. L11:1147 weakly suppresses
  10. L9:2381 weakly suppresses
  11. L7:2582 weakly excites
  12. L12:1060 weakly excites
  13. L10:2330 weakly suppresses
  14. L12:2592 weakly suppresses
  15. L10:2241 weakly excites
  16. L11:1055 weakly excites
  17. L12:2938 weakly excites
  18. L11:381 weakly suppresses
  19. L4:2252 weakly suppresses
  20. L7:2616 weakly excites
  21. L8:913 weakly suppresses
  22. L8:389 weakly suppresses
  23. L5:692 weakly suppresses
  24. L11:1532 weakly suppresses

Feeds later layers

  1. L15:1464 suppresses
  2. L26:825 suppresses
  3. L15:2354 suppresses
  4. L24:893 suppresses
  5. L19:1884 suppresses
  6. L23:476 suppresses
  7. L23:370 suppresses
  8. L24:116 excites
  9. L26:207 excites
  10. L27:405 suppresses
  11. L17:2515 suppresses
  12. L22:439 excites
  13. L14:2669 suppresses
  14. L24:1284 suppresses
  15. L26:584 suppresses
  16. L21:818 suppresses
  17. L26:786 excites
  18. L25:2282 suppresses
  19. L18:1977 suppresses
  20. L18:915 suppresses
  21. L19:241 excites
  22. L21:1279 suppresses unclear: function words in descriptive prose
  23. L21:628 suppresses
  24. L19:543 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 -2.86 and -2.691 values between -2.18 and -2.012 values between -2.01 and -1.853 values between -1.68 and -1.512 values between -1.51 and -1.345 values between -1.34 and -1.173 values between -1.17 and -1.005 values between -1.00 and -0.8311 values between -0.83 and -0.6626 values between -0.66 and -0.4977 values between -0.49 and -0.32366 values between -0.32 and -0.152040 values between -0.15 and 0.021106 values between 0.02 and 0.19157 values between 0.19 and 0.3547 values between 0.35 and 0.5223 values between 0.52 and 0.6913 values between 0.69 and 0.866 values between 0.86 and 1.034 values between 1.03 and 1.203 values between 1.20 and 1.371 values between 1.37 and 1.543 values between 1.54 and 1.712 values between 1.71 and 1.881 values between 3.74 and 3.91offnegativepositiveoften

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. 版
  2. uchi
  3. ized
  4. 绝
  5. icl
  6. ELY
  7. 집
  8. utt
  9. gtest
  10. 备

Suppresses

  1. 在香港
  2. 在网络
  3. 在里面
  4. 在学校
  5. 在深圳
  6. 在广州
  7. 在
  8. 在他
  9. 在当地
  10. 在这里