Neuron Explorer

Qwen3-0.6B

end of document marker token

A neuron in layer 10 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 the separator that marks the end of one web document and the start of another, especially after a complete story or paragraph. Turning it up did not visibly change the model's writing in a clear way.

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

    died and my father works abroad all year. I'm happy when my granddad isn't drunk, we have food to eat and it's warm at home."<|endoftext|>Memory Game - Charley Harper's Animals⏎From beagles and bobcats

  2. #2 peak

    Racial barriers failed to keep Bessie Coleman from taking her place in the sky. What might she have accomplished had she lived to a ripe old age?<|endoftext|>Sixteen years ago, Patricia L. Gerbarg, a Kingston, N

  3. #3 peak

    two cobras. She holds in her hands the sceptre and ankh. Many times she has the head of a lioness or is shown as one.<|endoftext|>More about the pharmacist. Notice the double entendre of⏎"tinctured

  4. #4 peak

    can benefit from signing.⏎"It's easier for them to grasp," said Skaja, an exceptional/gifted student educator with Brevard Public Schools. ...<|endoftext|>When planning your next landscape project, consider using native species. Native plants tend to

  5. #5 peak

    Six Jewish musicians from the philharmonic were murdered by the Nazis in Austria and 11 were deported to death camps, according to reports.<|endoftext|>Wednesday 29 June 2022⏎Photoelectrochemical cells

  6. #6 peak

    surface areas. Then a tripod scale with a winch is used to weigh the seal; females can bulk up to about 600 kilograms.⏎The scientists<|endoftext|>Parents often read to their children as a way to spend time together and

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: the relationship between the concept of "moral development" and the concept of "ethical philosophy". This is a broad topic that intersects in several ways, including the study of philosophy as a discipline, focusing on the interrelationships between ethics and politics

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

    visit any of the following sites:<|endoftext|>Most of us are addicted to our morning cup of tea to carry out the routine activities of the day. But have you considered replacing your morning cup of normal

  2. #2 peak

    , the last 16 years living in China, and no one had ever asked me that question.⏎As an investment adviser, with the perspective of a development economist, my life seems to have been an endless search for answers to the question of

  3. #3 peak

    words.⏎Professor Onur Güntürkün, one of the co-investigators from of Ruhr University’s Department of Biopsychology, says “that pigeons — separated by 300 million years

  4. #4 peak

    . We will explore the bromide in the Allegheny River and where it comes from (Figure

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. L9:368 weakly excites
  2. L2:763 weakly suppresses
  3. L3:2219 weakly excites
  4. L1:308 weakly suppresses
  5. L7:775 weakly suppresses
  6. L2:718 weakly suppresses
  7. L5:2547 weakly suppresses
  8. L7:276 weakly suppresses
  9. L4:1549 weakly excites
  10. L0:493 weakly excites
  11. L6:533 weakly suppresses
  12. L8:1356 weakly excites
  13. L1:1162 weakly excites
  14. L4:577 weakly suppresses
  15. L3:386 weakly suppresses
  16. L6:954 weakly excites
  17. L2:1416 weakly suppresses
  18. L8:254 weakly excites
  19. L2:890 weakly excites
  20. L0:2143 weakly suppresses
  21. L9:1212 weakly suppresses
  22. L6:2800 weakly excites
  23. L3:274 weakly excites
  24. L4:242 weakly excites

Feeds later layers

  1. L17:1333 weakly excites
  2. L11:2646 weakly excites
  3. L15:1871 weakly suppresses
  4. L15:272 weakly suppresses
  5. L14:331 weakly suppresses
  6. L12:2611 weakly suppresses
  7. L17:2091 weakly suppresses
  8. L18:2755 weakly excites
  9. L21:2372 weakly suppresses
  10. L13:3041 weakly excites
  11. L15:703 weakly suppresses
  12. L13:1661 weakly excites
  13. L14:2233 weakly excites
  14. L14:1234 weakly suppresses
  15. L14:2534 weakly suppresses
  16. L26:436 weakly excites
  17. L19:2865 weakly excites
  18. L12:633 weakly suppresses
  19. L19:2832 weakly suppresses
  20. L12:3021 weakly excites
  21. L11:2522 weakly suppresses
  22. L21:760 weakly excites
  23. L15:2055 weakly suppresses
  24. L18:2878 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 -1.15 and -1.051 values between -0.85 and -0.751 values between -0.75 and -0.657 values between -0.65 and -0.5512 values between -0.55 and -0.4536 values between -0.45 and -0.34142 values between -0.34 and -0.24337 values between -0.24 and -0.14809 values between -0.14 and -0.041603 values between -0.04 and 0.06592 values between 0.06 and 0.16238 values between 0.16 and 0.2682 values between 0.26 and 0.3629 values between 0.36 and 0.468 values between 0.46 and 0.561 values between 0.56 and 0.672 values between 0.67 and 0.771 values between 0.77 and 0.871 values between 0.87 and 0.971 values between 0.97 and 1.071 values between 1.07 and 1.171 values between 1.67 and 1.771 values between 2.78 and 2.88offnegativepositiveoften

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. .Go
  2. moc
  3. lov
  4. knockout
  5. awa
  6. ramid
  7. enerator
  8. jom
  9. 马桶
  10. جو

Suppresses

  1. рг
  2. ernes
  3. стр
  4. ethical
  5. 滞
  6. 会在
  7. 白
  8. 仅供参考
  9. -<?
  10. 思索