Neuron Explorer

Qwen3-0.6B

document ends and list line breaks

A neuron in layer 14 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 mostly reacts to the invisible marker that signals the end of a web page or document, and to line breaks in lists. It seems to care about where one piece of text stops and another starts. Turning it down made the model get stuck writing endless blank lines.

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

    , each of which is shown in turn below:⏎The distance between any grid point and the rectangular border is one unit of measurement (equal to the length of the<|endoftext|>Shree Ganeshay Namah⏎How to write Anchoring Script in English

  2. #2 peak

    Resources⏎- Managerial Economics - Discussion⏎- Selected Reading⏎- UPSC IAS Exams Notes⏎- Developer's Best Practices⏎- Questions and Answers⏎- Effective Resume Writing⏎- HR Interview Questions⏎- Computer Glossary⏎-

  3. #3 peak

    become an integral part of the semiconductor industry. Lithography is widely used in the industry to make two-dimensional circuit boards, but a convenient method for creating three-dimensional<|endoftext|>Embedding a Racial Equity Perspective in the Positive Youth Development Approach⏎This brief

  4. #4 peak

    using everyday materials.⏎Unit blocks are an excellent STEAM tool because they address Science, Technology (how things work), Engineering (problem solving) Art , and Math<|endoftext|>What is the maximum amount that sea levels could rise by 2100

  5. #5 peak

    therefore, is a time for focusing on the heart, a time for asking questions about our spiritual health:⏎- What are my characteristic sins, and how can I<|endoftext|>An ion-chromatographic method for the direct determination of ammonium, potassium

  6. #6 peak

    is a path that starts and ends at the same node. Sometimes cycles are useful, but they often represent challenges in graph algorithms. Failing to detect a cycle<|endoftext|>For hundreds of years men of science and theoretical physics have been struggling to solve the

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

    .NET is in the second position with only 20.3%. The statistics also shows that PHP is the sixth popular of all the programming language. (Source:⏎- Easier to learn and Implement⏎PHP is easier to learn than its competitors

  2. #2 peak

    of the University Community: (Students, faculty, employees,⏎and alumni.) The University of Miami Physics Department maintains an⏎Observatory for teaching

  3. #3 peak

    its story.⏎We've sent an email with your order details. Order ID #:⏎To access this title, visit your library in the app or on the desktop website

  4. #4 peak

    Clinical | Educational | Industrial | Professional items | World psychology |⏎Mindfulness-Based Cognitive Therapy (MBCT) is a method of therapy which blends features of two disciplines:⏎- Cognitive therapy aims to identify and alter cognitive distortions (warped or inaccurate

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. L10:84 strongly suppresses
  2. L13:10 excites
  3. L1:1279 suppresses
  4. L0:37 excites
  5. L5:331 suppresses
  6. L3:98 excites
  7. L6:67 excites
  8. L10:259 excites
  9. L9:25 excites
  10. L2:204 excites
  11. L10:33 suppresses
  12. L5:238 excites
  13. L0:596 excites
  14. L1:1 excites
  15. L6:216 excites
  16. L8:88 excites line breaks inside wrapped sentences
  17. L8:32 excites
  18. L9:7 excites
  19. L13:1721 suppresses
  20. L12:5 excites
  21. L9:2729 suppresses
  22. L0:1224 suppresses
  23. L0:178 suppresses
  24. L0:1301 suppresses

Feeds later layers

  1. L15:81 strongly suppresses line breaks inside running text
  2. L17:829 strongly suppresses
  3. L18:429 strongly suppresses
  4. L16:313 strongly suppresses
  5. L17:205 strongly suppresses
  6. L20:346 strongly suppresses
  7. L16:359 strongly suppresses
  8. L18:5 strongly suppresses
  9. L17:1073 suppresses
  10. L16:190 suppresses end-of-document boundary token
  11. L19:187 suppresses
  12. L18:705 suppresses
  13. L18:3042 suppresses line breaks in wrapped text
  14. L21:130 suppresses
  15. L16:1421 suppresses
  16. L17:921 suppresses
  17. L20:467 suppresses
  18. L16:1579 suppresses
  19. L18:886 suppresses
  20. L17:432 suppresses
  21. L15:186 suppresses
  22. L22:506 suppresses
  23. L16:414 suppresses
  24. L17:554 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 -7.88 and -7.663 values between -7.43 and -7.212 values between -7.21 and -6.982 values between -6.31 and -6.082 values between -6.08 and -5.865 values between -5.86 and -5.641 values between -5.64 and -5.413 values between -5.41 and -5.193 values between -5.19 and -4.964 values between -4.96 and -4.742 values between -4.74 and -4.515 values between -4.51 and -4.295 values between -4.29 and -4.075 values between -4.07 and -3.845 values between -3.84 and -3.622 values between -3.62 and -3.392 values between -3.39 and -3.174 values between -3.17 and -2.943 values between -2.94 and -2.726 values between -2.72 and -2.493 values between -2.49 and -2.273 values between -2.27 and -2.052 values between -2.05 and -1.822 values between -1.82 and -1.601 values between -1.37 and -1.152 values between -0.92 and -0.702 values between -0.70 and -0.482 values between -0.48 and -0.258 values between -0.25 and -0.03622 values between -0.03 and 0.201671 values between 0.20 and 0.421079 values between 0.42 and 0.65395 values between 0.65 and 0.8751 values between 0.87 and 1.10offnegativepositiveoften

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. sexes
  2. ANDOM
  3. 下班
  4. '${
  5. '@/
  6. .sex
  7. verses
  8. urry
  9. ptic
  10. `;⏎⏎

Suppresses

  1. illos
  2. illo
  3. span
  4. s
  5. black
  6. think
  7. 展开
  8. lette
  9. spl
  10. ships