Neuron Explorer

Qwen3-0.6B

comma before 'such as' examples

A neuron in layer 24 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 comma that comes right before a list of examples, such as 'other symptoms of appendicitis, such as pain'. Turning it up made the model repeat the word 'such' endlessly, showing it is tied to that kind of example phrasing.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    will be built.⏎The four basic techniques of ML are: supervised , unsupervised, semi-supervised ,and reinforcement learning. There are other Machine Learning techniques, so the below list is not exhaustive.⏎Supervised machine learning⏎This technique uses

  2. #2 peak

    be causing your symptoms.⏎Your doctor may also order other tests, including:⏎- Hysteroscopy. Using a viewing instrument inserted through the vagina

  3. #3 peak

    blood cells to fight the infection. The increase in white blood cells can lead to a fever. A fever may also be accompanied by other symptoms of appendicitis, such as pain, swelling, and loss of appetite.⏎Gas is another common symptom

  4. #4 peak

    (such as www.internetsociety.org) to Internet addresses that are used by computers to talk to each other.⏎DNS also provides other information about domain names, such as mail services.⏎Why is DNS important?⏎Domain names are fundamental to the

  5. #5 peak

    als or their derivatives, such as flour, starch, bread, pasta, biscuits and cakes, all contain gluten. Gluten may also be present in less obvious foods, such as sweets, puddings, preserved meats, stock cubes, etc.⏎What

  6. #6 peak

    made up of A, B, AB, or O blood types, and the Rh group is either positive or negative. You blood groups help doctors know many things, like what kind of blood you’ll need in a blood transfusion or an organ

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 importance of the larger human and global human and global human and human value and importance and importance and importance and importance as well as the importance of the importance of the importance of the importance of the importance of the importance... I like to like

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

    solutions, questions and answers of programming and scripting languages:<|endoftext|>RALEIGH — New research has found that North Carolina’s shellfish industry provides over $27 million in economic impact and 532 jobs in the state. Until 2

  2. #2 peak

    specificity of 25% to 80%. Typically a fracture can be seen on ultrasonography as a disruption in the hyperechoic linear cortex of bone as well as surrounding hematoma. (See image

  3. #3 peak

    time from a packed agenda to read aloud Boylan's request. The council members, struck by Boylen's youthful enthusiasm, passed the hat and collected $10.00.⏎Flush with funds, Boylan and Cook experimented with various rocket

  4. #4 peak

    sustainable eating and Angelina Frankowska (University of Manchester), whose expertise lays in sustainable foods and food groups.⏎The event attracted an audience of 70, including the members of the general public as well as staff and students from the University

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. L22:149 suppresses
  2. L19:2614 excites
  3. L20:2676 excites
  4. L17:2545 excites
  5. L19:226 excites
  6. L18:178 suppresses
  7. L15:2393 suppresses
  8. L19:1124 excites
  9. L13:1567 suppresses
  10. L20:728 excites
  11. L19:2918 weakly suppresses
  12. L23:1448 weakly suppresses
  13. L9:77 weakly excites commas in citations and name lists
  14. L19:158 weakly suppresses
  15. L14:704 weakly excites
  16. L23:1068 weakly suppresses
  17. L16:240 weakly excites comma after 'sometimes' or 'the truth'
  18. L13:2704 weakly excites
  19. L18:1438 weakly excites
  20. L20:593 weakly suppresses
  21. L20:2038 weakly suppresses
  22. L20:42 weakly excites
  23. L9:1165 weakly excites
  24. L22:974 weakly excites

Feeds later layers

  1. L27:694 strongly excites
  2. L25:1898 strongly excites
  3. L27:927 weakly suppresses
  4. L25:769 weakly excites
  5. L25:1909 weakly excites unclear: mid-word pieces of names
  6. L26:1777 weakly suppresses
  7. L26:2051 weakly excites
  8. L27:2895 weakly excites
  9. L27:1981 weakly excites
  10. L26:2767 weakly suppresses
  11. L25:1105 weakly suppresses
  12. L25:440 weakly suppresses
  13. L27:1953 weakly excites
  14. L27:759 weakly suppresses
  15. L25:2182 weakly suppresses
  16. L26:2486 weakly excites
  17. L25:95 weakly excites
  18. L25:780 weakly suppresses opening parenthesis before an acronym
  19. L26:1238 weakly excites
  20. L27:2785 weakly excites
  21. L25:1467 weakly excites
  22. L25:563 weakly suppresses
  23. L25:1804 weakly suppresses
  24. L25:2548 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 -26.00 and -24.551 values between -20.19 and -18.742 values between -17.29 and -15.835 values between -15.83 and -14.387 values between -14.38 and -12.936 values between -12.93 and -11.4815 values between -11.48 and -10.0228 values between -10.02 and -8.5747 values between -8.57 and -7.1285 values between -7.12 and -5.67140 values between -5.67 and -4.22171 values between -4.22 and -2.76345 values between -2.76 and -1.311566 values between -1.31 and 0.14896 values between 0.14 and 1.60219 values between 1.60 and 3.05135 values between 3.05 and 4.5096 values between 4.50 and 5.9551 values between 5.95 and 7.4037 values between 7.40 and 8.8620 values between 8.86 and 10.3110 values between 10.31 and 11.767 values between 11.76 and 13.214 values between 13.21 and 14.673 values between 14.67 and 16.121 values between 16.12 and 17.573 values between 19.02 and 20.483 values between 20.48 and 21.932 values between 21.93 and 23.382 values between 30.64 and 32.09offnegativepositiveoften

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. such
  2. such
  3. Such
  4. Such
  5. 这样的
  6. SUCH
  7. 這樣的
  8. 如此
  9. 此类
  10. 这样

Suppresses

  1. strup
  2. ertype
  3. ificance
  4. liği
  5. đen
  6. 看你
  7. amental
  8. ================================
  9. responseObject
  10. 臌