Neuron Explorer

Qwen3-0.6B

second item in a parallel pair or list

A neuron in layer 13 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron tends to fire on the second of two matching things, such as the second item in a list of parts, the second of a pair of words like 'one and two', or a qualifier like 'normal circumstances'. Turning it up made the model get stuck repeating phrases.

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

    Diode (x4)⏎- 1N5819 Schottky Diode⏎- Resistor (2MΩ, 560Ω, 1KΩ, 10Ω, 120Ω

  2. #2 peak

    life mener une vie normale⏎→ He has occasional injections to maintain his good health but otherwise he lives a normal life.⏎in normal circumstances, under normal circumstances en temps normal⏎→ In normal circumstances I would have resigned immediately.⏎to continue

  3. #3 peak

    educators establish a respectful classroom environment to avoid having students draw comparisons between the characters in the film and their classmates.⏎Before Watching: Consider whether you strongly agree, agree, disagree, or strongly disagree with this statement: Sometimes we have to change who we

  4. #4 peak

    -smallest county in Utah by land area and third-smallest by total area.⏎The county extends from high in the Wasatch Range in the east

  5. #5 peak

    cheesy things) to the delicious dish.⏎As the Spanish influence infiltrated the New World, other ingredients were added to the cheese quesadilla such as chicken, beef, turkey, seafood and vegetables. The quesadilla evolved into a blend of

  6. #6 peak

    in the elements. This is because the elements are listed in part by their electron configuration. The Alkali metals and Alkaline earth metals have one and two valence electrons (electrons in the outer shell) respectively. These elements lose

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: if I were to choose, if I had the if I I I were I I I if I I I if I I I I if I I I I if I I if I I I if I I I if I I if I

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

    . They can increase the power flux of sunlight hundreds of times.⏎Conductor — The material through which electricity is transmitted, such as an electrical wire, or transmission or distribution line.⏎Converter — A unit that converts a direct current (dc) voltage

  2. #2 peak

    ICS In THE unITED STATES Arcoxia, Vioxx, Acomplia, Zelnorm, and Tysabri; Aranesp, Procrit, Avandia, and Lucentis —all of these prominent

  3. #3 peak

    reading this topic⏎0 members, 0 guests, 0 anonymous users<|endoftext|>The JCTC applauds the action by the FDA to monitor e

  4. #4 peak

    grains should not come in contact with any metal object. This suggestion is made for both milk kefir-grains and water kefir-grains, or sugary kefir-grains [SKG]. This has become a firm suggestion with myth surrounding

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. L12:214 excites
  2. L12:964 excites
  3. L8:241 suppresses
  4. L12:2934 excites
  5. L10:1771 suppresses
  6. L12:1094 excites
  7. L12:798 excites
  8. L11:47 suppresses
  9. L12:836 suppresses
  10. L12:1720 suppresses
  11. L10:453 suppresses
  12. L12:113 suppresses
  13. L12:1918 excites
  14. L12:314 suppresses
  15. L12:185 excites
  16. L8:548 excites
  17. L10:901 suppresses
  18. L12:863 excites
  19. L11:1082 excites
  20. L9:293 suppresses
  21. L12:147 weakly suppresses
  22. L8:47 weakly excites
  23. L10:366 weakly excites
  24. L11:644 weakly suppresses

Feeds later layers

  1. L14:1644 strongly suppresses
  2. L14:992 excites
  3. L15:1705 excites
  4. L15:2205 suppresses
  5. L14:2292 excites
  6. L14:1370 suppresses
  7. L14:1773 excites
  8. L15:628 excites
  9. L14:2206 excites
  10. L14:2570 suppresses
  11. L14:1448 excites
  12. L15:2399 suppresses
  13. L14:695 excites
  14. L15:1660 excites 'or' in lists of alternatives
  15. L14:1569 excites
  16. L14:816 excites
  17. L14:2754 suppresses
  18. L15:150 excites
  19. L14:846 excites
  20. L15:2287 excites
  21. L14:1524 excites
  22. L14:1781 excites
  23. L14:900 excites
  24. L16:1774 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 -4.30 and -4.091 values between -2.84 and -2.631 values between -2.22 and -2.012 values between -2.01 and -1.801 values between -1.80 and -1.602 values between -1.60 and -1.3910 values between -1.39 and -1.186 values between -1.18 and -0.9710 values between -0.97 and -0.7637 values between -0.76 and -0.5697 values between -0.56 and -0.35620 values between -0.35 and -0.142326 values between -0.14 and 0.07586 values between 0.07 and 0.2880 values between 0.28 and 0.4836 values between 0.48 and 0.6932 values between 0.69 and 0.9018 values between 0.90 and 1.1115 values between 1.11 and 1.316 values between 1.31 and 1.527 values between 1.52 and 1.735 values between 1.73 and 1.942 values between 1.94 and 2.152 values between 2.15 and 2.352 values between 2.56 and 2.771 values between 2.98 and 3.181 values between 3.18 and 3.391 values between 3.81 and 4.02offnegativepositiveoften

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. ohl
  2. Richt
  3. Guides
  4. chiar
  5. avage
  6. appliances
  7. resses
  8. Appliances
  9. ections
  10. rue

Suppresses

  1. 岁时
  2. etas
  3. beta
  4. -backed
  5. bac
  6. beta
  7. 多年前
  8. idebar
  9. ッチ
  10. osaic