Neuron Explorer

Qwen3-0.6B

end-of-text marker at document breaks

A neuron in layer 4 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 reacts to the hidden marker the model uses to show that one document has ended and another begins. It is mostly silent otherwise. Turning it up did not produce a clear, consistent change in what the model wrote.

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

    , 2012<|endoftext|>Some weather conditions are simply not fun to ride in. As cyclists, windy days

  2. #2 peak

    0240.<|endoftext|>By 2003, plenty of data suggested that pesticide drift was harming

  3. #3 peak

    ,0<|endoftext|>Guidelines for Restoring Cultural Landscapes⏎Identifying, retaining and preserving

  4. #4 peak

    rounded to the nearest integer.<|endoftext|>Christine Duncan, MD⏎Sickle cell disease (SCD) affects as

  5. #5 peak

    go<|endoftext|>A running scroll of past classes I have offered……..⏎DEFINITION: The

  6. #6 peak

    718<|endoftext|>Women who have kids later on in life may live longer, according to the findings

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 user's name is not the same as the letter, and the user's name is not the same as the letter. Let's see the image (or not) (the image) or not the image. Let's see the image

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

    position of formal authority. Alternatively then, if someone without a position of formal leadership is attempting to dominate the conversation, other team members may be less likely to acquiesce.⏎The above hypotheses were tested by Tost, G

  2. #2 peak

    is, in the United States.⏎The national Constitution enumerates the powers which may be exercised by the national government, and reserves all other powers “to the States respectively, or to the people.” Because of this dual or double character of our system

  3. #3 peak

    1x1, 2x2, 3x3, skill games) and 5x5 play, trying not<|endoftext|>Botanical Name: Asclepias exaltata⏎Plant Family: Milkweed family (Asclepi

  4. #4 peak

    transferred to Punjab;⏎iv) demand relating to halting the diversion of Punjab's river water.⏎There was a growing frustration as the Centre was not ready to acquiesce to their political demands. This provided a route to the dharma yud

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. L3:236 excites
  2. L3:421 weakly suppresses
  3. L3:50 weakly excites
  4. L2:75 weakly suppresses
  5. L2:2328 weakly suppresses
  6. L3:189 weakly excites
  7. L1:2005 weakly excites
  8. L1:951 weakly excites
  9. L3:464 weakly excites
  10. L3:474 weakly excites
  11. L2:1920 weakly suppresses
  12. L1:1606 weakly excites
  13. L3:2850 weakly suppresses
  14. L3:2140 weakly suppresses
  15. L3:456 weakly suppresses
  16. L3:134 weakly suppresses
  17. L3:186 weakly suppresses
  18. L3:832 weakly suppresses
  19. L3:1594 weakly excites
  20. L2:1494 weakly suppresses
  21. L3:262 weakly suppresses
  22. L2:2788 weakly suppresses
  23. L1:1438 weakly excites
  24. L2:112 weakly excites

Feeds later layers

  1. L5:471 excites
  2. L5:11 excites
  3. L8:1365 excites
  4. L6:49 suppresses
  5. L9:81 excites
  6. L6:730 excites
  7. L18:2003 excites end-of-document separator token
  8. L10:507 excites
  9. L17:2580 weakly suppresses
  10. L10:227 weakly suppresses middle of fixed names and phrases
  11. L10:2957 weakly suppresses
  12. L8:1186 weakly excites
  13. L17:1751 weakly suppresses
  14. L5:1196 weakly suppresses
  15. L15:794 weakly suppresses
  16. L18:860 weakly excites
  17. L5:315 weakly excites
  18. L16:1709 weakly suppresses
  19. L7:105 weakly excites
  20. L13:2264 weakly excites
  21. L11:2341 weakly suppresses
  22. L9:1048 weakly suppresses
  23. L9:2880 weakly excites
  24. L13:283 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.

7 values between -0.62 and -0.5318 values between -0.53 and -0.4438 values between -0.44 and -0.35115 values between -0.35 and -0.26301 values between -0.26 and -0.17828 values between -0.17 and -0.082000 values between -0.08 and 0.01499 values between 0.01 and 0.0979 values between 0.09 and 0.1819 values between 0.18 and 0.273 values between 0.27 and 0.361 values between 2.84 and 2.93offnegativepositiveoften

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. hydration
  2. kickoff
  3. modeling
  4. lient
  5. PRESSION
  6. pects
  7. orian
  8. illac
  9. signature
  10. implications

Suppresses

  1. owered
  2. rays
  3. blick
  4. viz
  5. ưỡ
  6. ruta
  7. atient
  8. .TRUE
  9. FXML
  10. дал