Neuron Explorer

Qwen3-0.6B

'a' after verbs like require or without

A neuron in layer 20 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 word "a" when it comes right after a verb or a word like "without" or "require", such as "without a clear path" or "requires a level of precision". Turning it up didn't produce anything that clearly matches this pattern, so the steering results don't add much.

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

    standards for washability, mold and microbial resistance, and fire safety. Ceiling panels with these characteristics are ideal for patient rooms, treatment rooms and other spaces that require a balance of cleanliness and quiet.⏎Ceiling panels with a NRC of 0

  2. #2 peak

    PV works year round. It uses daylight to generate electricity. Naturally, on rainy days or shorter winter days, there is less daylight so the system will not produce a lot of power. Call to our office to see our fully functioning solar PV system

  3. #3 peak

    aver in an emergency. If there is a fire in your home, for example, a single stairwell may become unusable, leaving family members and tenants without a clear path to safety. In basements, egress windows are essential.⏎An

  4. #4 peak

    When women were accorded full status as electors in August 1920 by the Nineteenth Amendment to the federal Constitution, the...⏎Users without a subscription are not able to see the full⏎to access all content.<|endoftext|>Great

  5. #5 peak

    freedom they deserved. The slaves were like puppets that the American people could control. The slaves couldn’t vote, they didn’t have a choice on what they were going to do that day, and they couldn’t say

  6. #6 peak

    breathing, while vocal cord surgery addresses similar problems with sound and articulation.⏎In many cases, surgeons operate on delicate structures such as minute blood vessels, which requires a level of precision and expertise that our medical staff are skilled at providing. Common procedures

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: I am a new software development software and I have been told that software test software test software test software test software test software test software test software test software test software test software test software test software test software test software test software test software test software

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

    help you fall asleep, too much will have you “ping” awake in the middle of the night. This will drastically affect your quality of sleep.⏎- Create total darkness and quiet

  2. #2 peak

    Mitanni was a multiethnic empire composed of different kingdoms spread out through the land between Anatolia (modern Turkey) and Babylon (modern Iraq).⏎Mitanni took brief control of the region in the Tigris and Euphrates basin between

  3. #3 peak

    is Noether's theorem, which states that there is a one-to-one correspondence between each one of them and a differentiable symmetry of

  4. #4 peak

    to force tax hikes onto local residents in an effort to bring funding for poorer school districts into greater parity with wealthier districts. It was at this point that the concept of self

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. L18:487 excites
  2. L19:2406 excites
  3. L18:934 suppresses
  4. L18:2815 suppresses
  5. L17:2009 suppresses
  6. L18:2720 suppresses
  7. L18:1498 suppresses
  8. L18:1084 excites
  9. L18:861 excites
  10. L15:2133 suppresses
  11. L18:1359 weakly suppresses
  12. L8:336 weakly excites
  13. L17:2659 weakly suppresses
  14. L3:1218 weakly suppresses
  15. L18:178 weakly suppresses
  16. L19:2761 weakly suppresses
  17. L19:262 weakly excites
  18. L10:155 weakly suppresses
  19. L17:2685 weakly suppresses
  20. L19:1180 weakly excites
  21. L16:513 weakly suppresses
  22. L10:783 weakly suppresses
  23. L17:2914 weakly suppresses
  24. L1:689 weakly suppresses

Feeds later layers

  1. L24:153 suppresses
  2. L21:1501 excites
  3. L22:1081 excites
  4. L22:496 weakly suppresses
  5. L27:434 weakly excites
  6. L23:1535 weakly excites
  7. L22:211 weakly excites
  8. L25:390 weakly excites
  9. L25:284 weakly suppresses
  10. L21:675 weakly suppresses
  11. L26:174 weakly suppresses
  12. L25:425 weakly suppresses
  13. L27:1102 weakly excites
  14. L27:814 weakly excites
  15. L21:2557 weakly excites
  16. L22:626 weakly suppresses
  17. L26:678 weakly suppresses
  18. L22:1881 weakly excites
  19. L25:410 weakly excites
  20. L23:1529 weakly excites
  21. L26:2099 weakly excites
  22. L21:758 weakly excites
  23. L23:218 weakly excites
  24. L27:556 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 -14.08 and -13.061 values between -13.06 and -12.051 values between -12.05 and -11.031 values between -11.03 and -10.022 values between -9.00 and -7.999 values between -7.99 and -6.9713 values between -6.97 and -5.9611 values between -5.96 and -4.9413 values between -4.94 and -3.9320 values between -3.93 and -2.9147 values between -2.91 and -1.90175 values between -1.90 and -0.882439 values between -0.88 and 0.14962 values between 0.14 and 1.1584 values between 1.15 and 2.1734 values between 2.17 and 3.1823 values between 3.18 and 4.2015 values between 4.20 and 5.2114 values between 5.21 and 6.238 values between 6.23 and 7.2413 values between 7.24 and 8.267 values between 8.26 and 9.274 values between 9.27 and 10.291 values between 10.29 and 11.302 values between 12.32 and 13.331 values between 13.33 and 14.353 values between 14.35 and 15.361 values between 15.36 and 16.381 values between 17.39 and 18.411 values between 19.43 and 20.441 values between 25.52 and 26.53offnegativepositiveoften

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. sense
  2. approach
  3. sense
  4. 度
  5. knack
  6. 一场
  7. 률
  8. basis
  9. standpoint
  10. degree

Suppresses

  1. øj
  2. ighbors
  3. olutely
  4. ног
  5. dotyczące
  6. payments
  7. 鸨
  8. ây
  9. anz
  10. miejscowości