Neuron Explorer

Qwen3-0.6B

unclear: mid-sentence words before a continuation

A neuron in layer 17 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 fires on a grab bag of ordinary words that sit in the middle of a sentence, such as 'will', 'is', 'and' and 'for', where the sentence clearly has to keep going. The examples do not share one clear pattern, so treat it as hard to pin down. Turning it up made the model drift into generic or rambling text.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).

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

    to the Eagles' "Hotel California"—Elaine Treharne and Claude Willan offer a clear yet nuanced overview of how humans convey meaning. Text Technologies will enable students and teachers to generate multiple lines of inquiry into how communication—its production

  2. #2 peak

    string in single quotes instead of double quotes if you want to print double quotes in a string:⏎print(‘John says “Hello there!”’)⏎John says “Hello there!”⏎## Use escape sequences to print special characters, tabs, or

  3. #3 peak

    . Be sure to point out that a dead body feels no pain.⏎Let your child know that these cremated remains are placed in a⏎container called an urn and returned to the family. If cremation⏎has already taken place and the container

  4. #4 peak

    . The various tissue layers of the stomach wall then combine their functions to digest the bolus into a viscous, pulpy fluid called chyme. Chyme is directed into the duodenum<|endoftext|>Humans are social creatures, we were made to

  5. #5 peak

    crase in your small intestine assists with the breakdown of sucrose into fructose and glucose. This allows your body to absorb them, transport them to the liver for processing and distribute them throughout the body. The hormone insulin then facilitates the uptake of

  6. #6 peak

    so that they have a general idea of the appropriate colors. Then cut out the animals the children have been drawing and attach them to the habitats. The finished product is an attractive image of a wildlife scene that can teach the children a great deal about

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: this is a *possible* *possible* *possible* *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible *possible

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

    information. Citing a resource means including this specific information about the source in your work. Resources are cited using your instructor's choice of documentation styles.⏎What's a "documentation style"?⏎A documentation style is a standard, agreed-on method for

  2. #2 peak

    prototyping shows people what the feature will look like. But the code base or project is not necessarily used for the production version of the application.⏎What is a mockup logo?⏎What Is A Logo Mockup? A logo mockup is

  3. #3 peak

    .⏎- they took pride in the history and traditions of their people⏎- they organized forces to win independence⏎what happened in the russo-japanese war and what was the outcome?⏎- japan

  4. #4 peak

    16. A year later, the French settled in the area just north of present Montogm

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. L16:244 excites
  2. L15:2535 excites
  3. L10:28 suppresses unclear: weak, articles before nouns
  4. L10:150 suppresses
  5. L11:9 excites
  6. L11:1376 weakly suppresses
  7. L5:25 weakly excites
  8. L5:36 weakly suppresses
  9. L16:706 weakly suppresses
  10. L8:2653 weakly suppresses
  11. L9:3054 weakly suppresses
  12. L6:2342 weakly excites
  13. L9:15 weakly excites
  14. L16:445 weakly excites
  15. L8:2246 weakly suppresses
  16. L13:841 weakly excites
  17. L15:1944 weakly excites
  18. L13:2363 weakly excites
  19. L10:3064 weakly excites
  20. L15:2189 weakly excites
  21. L5:922 weakly excites
  22. L7:2054 weakly excites
  23. L15:76 weakly suppresses
  24. L10:765 weakly suppresses

Feeds later layers

  1. L18:1028 weakly suppresses
  2. L24:908 weakly suppresses
  3. L25:2518 weakly excites
  4. L22:1137 weakly excites
  5. L20:446 weakly suppresses
  6. L26:2948 weakly suppresses
  7. L27:2512 weakly suppresses
  8. L21:2637 weakly excites
  9. L21:1546 weakly excites
  10. L19:37 weakly suppresses
  11. L24:1066 weakly suppresses
  12. L18:2565 weakly suppresses
  13. L26:597 weakly excites
  14. L26:438 weakly excites
  15. L24:224 weakly excites
  16. L22:899 weakly suppresses
  17. L18:2701 weakly suppresses
  18. L22:97 weakly suppresses
  19. L21:2799 weakly suppresses
  20. L23:104 weakly excites
  21. L27:1120 weakly excites
  22. L22:2219 weakly suppresses
  23. L22:1543 weakly excites
  24. L21:2060 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.

2 values between -1.85 and -1.764 values between -1.76 and -1.673 values between -1.58 and -1.494 values between -1.40 and -1.305 values between -1.21 and -1.127 values between -1.12 and -1.036 values between -1.03 and -0.9414 values between -0.94 and -0.8414 values between -0.84 and -0.7517 values between -0.75 and -0.6623 values between -0.66 and -0.5737 values between -0.57 and -0.4872 values between -0.48 and -0.39103 values between -0.39 and -0.29195 values between -0.29 and -0.20301 values between -0.20 and -0.11709 values between -0.11 and -0.021348 values between -0.02 and 0.07473 values between 0.07 and 0.17252 values between 0.17 and 0.26140 values between 0.26 and 0.3572 values between 0.35 and 0.4436 values between 0.44 and 0.5324 values between 0.53 and 0.6315 values between 0.63 and 0.7214 values between 0.72 and 0.815 values between 0.81 and 0.902 values between 0.90 and 0.994 values between 0.99 and 1.083 values between 1.08 and 1.181 values between 1.18 and 1.271 values between 1.27 and 1.361 values between 1.64 and 1.731 values between 1.73 and 1.82offnegativepositiveoften

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. çon
  2. ifax
  3. actory
  4. ularity
  5. Lamar
  6. phy
  7. ạch
  8. angular
  9. atedRoute
  10. CDATA

Suppresses

  1. eva
  2. 金字
  3. vazgeç
  4. erview
  5. 不出来
  6. 家都知道
  7. 库里
  8. 成果转化
  9. 脱颖而出
  10. ihtiyacı