Neuron Explorer

Qwen3-0.6B

unclear: mixed, end-of-text and function words

A neuron in layer 15 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 grab bag of spots: the end of a document, words like 'from', 'which' and 'could', and ordinary places in sentences. No single theme stands out. When pushed up, the model's writing turned repetitive and wandered toward generic phrases, but that doesn't point to a clear meaning.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    were based on NASA’s lunar architecture plans.⏎The game itself only takes around 20 minutes to play, and despite its sophisticated 3D visuals—the game<|endoftext|>Disk or partition alignment is something new to me. I have heard of it before

  2. #2 peak

    for which mankind are warranted, individually or collectively in interfering with the liberty of action of any of their number, is self-protection. That the only purpose for which power can be rightfully exercised over

  3. #3 peak

    of Publication⏎Blanks, C. M. (1948). San Antonio the Inland Gate. Retrieved from https://digitalcommons.pvamu.edu/pvamu-theses/39

  4. #4 peak

    insects in their diet. Further down the line, he hopes these mini-cricket factories will catch on as a sustainable local food source for North American cities. "<|endoftext|>Philadelphia is the largest city in the Commonwealth of Pennsylvania, the second largest city on

  5. #5 peak

    $50 from the mint."<|endoftext|>In the next of our blog posts for Census week, we showcase recent research undertaken

  6. #6 peak

    than €1 each to produce. 10 confirmed patients have been connected to a breathing machine via a 3D printed valve.⏎Although Fracassi is<|endoftext|>Key: "S:" = Show Synset (semantic) relations, "W

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: (other than the other...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...)(...

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

    or not and take the default route. Here’s an example of⏎mwhen for Monoids:⏎-- Important numbers!⏎xs = [6, 9, 13,

  2. #2 peak

    , this function is only declared as:⏎void * memchr ( const void *, int, size_t );⏎instead of the two overloaded versions provided in C++.⏎/* memchr example */⏎int main ()⏎char * pch;⏎char str = "

  3. #3 peak

    Ones. For this deed, she went down in Assassin history as the legendary Egyptian Assassin who killed Cleopatra with an asp.⏎- 1 Biography⏎- 2 Legacy⏎- 3 Personality and traits⏎- 4 Behind the

  4. #4 peak

    with <BODY> and end with </BODY>, but in practice this isn't essential. The

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. L14:1609 weakly excites
  2. L8:2748 weakly suppresses
  3. L3:1951 weakly excites
  4. L12:2329 weakly excites
  5. L3:955 weakly excites
  6. L10:60 weakly excites
  7. L7:1805 weakly suppresses
  8. L2:853 weakly excites
  9. L8:1455 weakly suppresses
  10. L8:1513 weakly excites
  11. L6:477 weakly excites
  12. L0:37 weakly excites
  13. L6:2355 weakly suppresses
  14. L14:781 weakly excites
  15. L2:1205 weakly excites
  16. L7:1348 weakly suppresses
  17. L13:1223 weakly excites
  18. L7:1806 weakly excites
  19. L0:3039 weakly suppresses
  20. L14:777 weakly excites
  21. L2:590 weakly excites
  22. L2:1134 weakly suppresses
  23. L1:617 weakly excites
  24. L7:2205 weakly excites

Feeds later layers

  1. L16:1119 weakly excites
  2. L27:834 weakly suppresses
  3. L17:1342 weakly suppresses
  4. L22:1018 weakly excites
  5. L19:1518 weakly suppresses
  6. L23:1498 weakly suppresses
  7. L17:1351 weakly suppresses
  8. L20:1312 weakly excites
  9. L18:2483 weakly excites
  10. L18:2411 weakly excites
  11. L27:1375 weakly suppresses
  12. L18:1052 weakly excites
  13. L18:2925 weakly suppresses
  14. L16:2849 weakly suppresses
  15. L25:620 weakly excites
  16. L23:347 weakly excites
  17. L21:2699 weakly suppresses
  18. L16:2196 weakly suppresses
  19. L18:837 weakly suppresses
  20. L19:707 weakly excites
  21. L23:2113 weakly suppresses
  22. L18:1845 weakly suppresses
  23. L24:668 weakly suppresses
  24. L27:841 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 -1.28 and -1.171 values between -0.82 and -0.704 values between -0.70 and -0.596 values between -0.59 and -0.4768 values between -0.47 and -0.36261 values between -0.36 and -0.24669 values between -0.24 and -0.121218 values between -0.12 and -0.011140 values between -0.01 and 0.11309 values between 0.11 and 0.2385 values between 0.23 and 0.3442 values between 0.34 and 0.4627 values between 0.46 and 0.5717 values between 0.57 and 0.6912 values between 0.69 and 0.8113 values between 0.81 and 0.926 values between 0.92 and 1.046 values between 1.04 and 1.151 values between 1.15 and 1.275 values between 1.27 and 1.392 values between 1.39 and 1.501 values between 1.50 and 1.621 values between 1.62 and 1.732 values between 1.73 and 1.852 values between 1.85 and 1.972 values between 1.97 and 2.083 values between 2.20 and 2.311 values between 2.55 and 2.662 values between 2.66 and 2.781 values between 3.24 and 3.36offnegativepositiveoften

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. inger
  2. ude
  3. ypes
  4. aleza
  5. ube
  6. ylan
  7. .enable
  8. __;
  9. Alpha
  10. enable

Suppresses

  1. ronic
  2. цикл
  3. athi
  4. 衷
  5. loo
  6. 更大
  7. PROFITS
  8. wider
  9. broader
  10. rait