Neuron Explorer

Qwen3-0.6B

'a' or 'of' right after a verb

A neuron in layer 5 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 small word "a" (or "another", "of") when it comes straight after a verb, like "share a", "making a" or "preserve a", right before the object is named. Turning it up made the text repeat filler words, so the steering doesn't reveal much more.

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

    comes with wonderful side effects.⏎A transition to plant-based diets on the individual level, and away from animal agriculture on the societal level, will not only help prevent another pandemic — it will improve your health, spare animals from suffering, reduce environmental degradation

  2. #2 peak

    spaceflight program lacks a clear goal, policymakers and the public will say that NASA is adrift, even if astrophysicists and planetary scientists continue launching a panoply of missions. Moreover, scientists suffer programmatic whiplash whenever

  3. #3 peak

    posing this question, hoping that all your students take a moment to think about it, then hearing from the handful of students who have the time and courage to share a response, suppose you ask all of your students to respond to it using their click

  4. #4 peak

    . Some of the activities include making a jester hat, learning the chivalric code, making a shield, making

  5. #5 peak

    Elizabeth Williams, associate planner for the city.⏎RELATED: UW-Oshkosh Commencement rescheduled⏎Shoreline restoration includes efforts to preserve or restore a shoreline to its original state by planting vegetation that filters runoff, according to the Wisconsin

  6. #6 peak

    congestion we are currently experiencing will take several more months to clear out.⏎Unfortunately, we cannot tell you that those experts are wrong. What we can do is share a few foundational concepts from supply chain management that will help you better understand the port problem

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: " * * " " " " " " " " " " " " " " " " " " " " "

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

    attempt at loom building - before<|endoftext|>Did NASA's Curiosity Mars rover spot a flying saucer? (+video)⏎It sure did, but this was no UFO. The rover

  2. #2 peak

    family home is outdated and stuffy. The prairie, vast and empty, stretches away on every side.⏎Dreams end and realism begins when Carol takes a thirty-minute walk, inspecting the town, north and south, east and west.

  3. #3 peak

    certain dwarves, Fjalarr and Galarr, they called him into privy converse with them, and killed him, letting his blood run into two vats and a kettle. The kettle is named �

  4. #4 peak

    myth a means to express faith in the idea of progress. Even the old gods must admit that their successors are more beautiful and therefore better fitted to rule.⏎Book 1 opens in media res with the deposed god Saturn mourning the loss of his

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. L4:298 strongly suppresses
  2. L4:392 strongly suppresses
  3. L4:450 strongly suppresses
  4. L2:149 strongly excites
  5. L3:654 suppresses article 'a' after 'here are'
  6. L2:1426 suppresses
  7. L2:502 suppresses 'other' after a preposition
  8. L4:597 suppresses
  9. L4:705 suppresses
  10. L1:683 suppresses
  11. L3:858 suppresses
  12. L3:2559 suppresses
  13. L3:491 excites
  14. L4:740 suppresses
  15. L2:2177 excites
  16. L2:326 excites
  17. L2:568 excites
  18. L4:1052 suppresses
  19. L3:19 excites
  20. L2:579 suppresses
  21. L4:615 excites
  22. L4:2236 suppresses
  23. L1:427 excites
  24. L3:344 excites

Feeds later layers

  1. L8:498 strongly excites
  2. L6:415 strongly suppresses
  3. L11:793 strongly excites
  4. L9:543 strongly excites
  5. L13:1647 excites
  6. L7:636 suppresses
  7. L10:1190 excites
  8. L13:1675 excites
  9. L11:1684 suppresses
  10. L6:542 suppresses
  11. L6:457 suppresses
  12. L6:1682 suppresses
  13. L11:821 excites
  14. L15:2553 excites
  15. L6:90 excites
  16. L6:450 excites
  17. L15:2594 suppresses
  18. L13:1386 excites
  19. L13:1183 suppresses
  20. L6:307 excites
  21. L12:642 suppresses
  22. L18:1686 excites
  23. L14:2704 suppresses
  24. L10:246 suppresses

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 -2.26 and -2.122 values between -1.85 and -1.721 values between -1.58 and -1.454 values between -1.45 and -1.317 values between -1.31 and -1.186 values between -1.18 and -1.0411 values between -1.04 and -0.9125 values between -0.91 and -0.7721 values between -0.77 and -0.6344 values between -0.63 and -0.50135 values between -0.50 and -0.36425 values between -0.36 and -0.23764 values between -0.23 and -0.091729 values between -0.09 and 0.04405 values between 0.04 and 0.18119 values between 0.18 and 0.3169 values between 0.31 and 0.4534 values between 0.45 and 0.5823 values between 0.58 and 0.7216 values between 0.72 and 0.857 values between 0.85 and 0.9914 values between 0.99 and 1.135 values between 1.13 and 1.2612 values between 1.26 and 1.407 values between 1.40 and 1.535 values between 1.53 and 1.678 values between 1.67 and 1.802 values between 1.80 and 1.942 values between 2.07 and 2.212 values between 2.48 and 2.611 values between 2.61 and 2.752 values between 3.02 and 3.15offnegativepositiveoften

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. iera
  2. ual
  3. iche
  4. ouble
  5. UAL
  6. setData
  7. udio
  8. вен
  9. 又一次
  10. zos

Suppresses

  1. cripts
  2. CString
  3. ');</
  4. fälle
  5. 陆续
  6. ał
  7. ombat
  8. tlement
  9. ước
  10. различных