Neuron Explorer

Qwen3-0.6B

end of document boundary

A neuron in layer 17 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

The neuron fires on the invisible 'end of text' marker that separates one web page from the next, right after a finished document. Its negative side is ordinary sentence endings in the middle of text. Turning it up did not produce a clear theme.

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

    if you need to, you can always buy college assignments online.<|endoftext|>Breaking research news from a team of scientists led by Carnegie's Ho-kwang "

  2. #2 peak

    education after they reach puberty. “Many parents don’t want men to teach their daughters,” said Pushpa Kumar, the Okhla’s center’s teacher.<|endoftext|>When we begin to talk about "deferred" items in relation to accounting transactions

  3. #3 peak

    civil administration, but for ecclesiastical purposes remained part of Aker's parish. Aker borders Bymarken on the west side of the......<|endoftext|>Tell Time (Grade 2)⏎Videos and songs to help Grade 2 students

  4. #4 peak

    .⏎For severe low blood sugar, your child may need a glucagon injection. Ask your child’s healthcare provider about glucagon emergency kits for home and school.<|endoftext|>|This article needs additional citations for verification. (July 2012

  5. #5 peak

    ”? Not at all, says Wolf. Instead, she argues, “The great challenge now is to learn how to use both print and digital mediums to their best<|endoftext|>- A Tale of Two Brothers: Haiti’s Other Revolutions⏎Sévère

  6. #6 peak

    the 1950s is strongly indicated. At that time, colour photography existed, but was not as widely popular and affordable as black-and-white.<|endoftext|>Classification

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 problem of the country's issues of the time (i.e., the problems that are not known, not known at all at the time, that are not known at all at the time, not known at all at the time at the

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

    the date he received his first ever job (September 26, 1855) every year and he considered the date more important than his birthday. “All my future seemed to hinge on that day,” he reminisced later in

  2. #2 peak

    ⏎Create Volcano (or Atoll) martix. It is an extension of Bullseye matrix problem.⏎Who has power to do everything in this world?⏎Close MATLAB with keyboard⏎Find the treasures in MATLAB Central and discover how the community

  3. #3 peak

    vanizing and when to use one or the other. Or even both!⏎Contact Minnesota Industrial Coatings when you’re ready to apply America’s best industrial coating solutions. We provide our customers with consistently high-quality industrial coatings.⏎Galvanizing can be

  4. #4 peak

    from the American College of Physicians are important steps towards increasing access to and availability of the data necessary for measuring equity and inequity across and within all demographic groups. The authors call for future research to unpack the complex web of factors driving health and

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. L16:235 strongly suppresses
  2. L16:2524 strongly excites
  3. L15:2544 suppresses
  4. L16:2510 excites
  5. L15:101 excites
  6. L9:1039 excites
  7. L15:774 excites
  8. L13:969 suppresses
  9. L12:2927 suppresses
  10. L10:2423 suppresses
  11. L15:1246 suppresses
  12. L14:242 suppresses
  13. L13:2022 excites
  14. L10:1298 excites
  15. L15:2249 excites
  16. L2:1385 weakly excites
  17. L16:2046 weakly excites
  18. L16:717 weakly excites
  19. L14:2827 weakly suppresses
  20. L16:3064 weakly excites
  21. L16:2200 weakly excites
  22. L15:2675 weakly excites
  23. L16:2575 weakly excites
  24. L15:3020 weakly excites

Feeds later layers

  1. L18:2322 strongly excites end-of-text token after a finished passage
  2. L19:2057 strongly excites
  3. L26:2128 excites decimal points inside numbers
  4. L19:1069 excites
  5. L27:2888 weakly suppresses
  6. L26:1614 weakly suppresses
  7. L25:808 weakly excites
  8. L19:1527 weakly excites
  9. L23:1220 weakly excites
  10. L20:338 weakly excites
  11. L26:1118 weakly excites
  12. L22:417 weakly excites period closing a web blurb
  13. L21:1150 weakly suppresses
  14. L21:301 weakly suppresses
  15. L25:2518 weakly excites
  16. L18:1891 weakly suppresses
  17. L20:44 weakly suppresses
  18. L20:1125 weakly excites
  19. L18:1082 weakly excites
  20. L26:2197 weakly suppresses
  21. L25:269 weakly excites
  22. L26:16 weakly excites
  23. L26:319 weakly excites
  24. L26:449 weakly 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 -2.81 and -2.602 values between -2.40 and -2.192 values between -2.19 and -1.981 values between -1.57 and -1.361 values between -1.36 and -1.163 values between -1.16 and -0.958 values between -0.95 and -0.7414 values between -0.74 and -0.53151 values between -0.53 and -0.33951 values between -0.33 and -0.122033 values between -0.12 and 0.09637 values between 0.09 and 0.2985 values between 0.29 and 0.5010 values between 0.50 and 0.711 values between 1.12 and 1.331 values between 1.33 and 1.532 values between 1.74 and 1.951 values between 2.98 and 3.191 values between 3.39 and 3.601 values between 4.22 and 4.431 values between 5.05 and 5.251 values between 5.25 and 5.46offnegativepositiveoften

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. <|endoftext|>
  2. uces
  3. 功
  4. regards
  5. rel
  6. 哚
  7. ewn
  8. ただ
  9. ://
  10. ↩

Suppresses

  1. cul
  2. /OR
  3. QUICK
  4. ainers
  5. öst
  6. ROID
  7. èmes
  8. ashes
  9. Habitat
  10. anny