Neuron Explorer

Qwen3-0.6B

comma inside big numbers

A neuron in layer 22 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 in large numbers like 30,000 or 3,125. When it was pushed down, the model fell into loops of code-like symbols and blank lines, and when pushed up it wandered into vague prose.

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

    like for example snakes as non-vision. Therefore snakes using their pit organs, which is sensitive to EM between 5000nm and 30,000nm (IR), do not "see", but somehow "feel

  2. #2 peak

    , have you ever wondered how many academic journals are there in the world? Well, I do.⏎It is estimated that there are currently more than 30,000 academic journals, and the number continues to increase by about 5

  3. #3 peak

    calculation:⏎[(4,000 lbs + 6,000 lbs) × 1.25] ÷ 4 = 3,125 lbs⏎According to this calculation, the vehicle mentioned above will require

  4. #4 peak

    that you are ready to share it. First a few important points:⏎Now you have the modification as a source file. Sending the source file will not be helpful, as probably someone else has done other modifications to the original file in the meantime.

  5. #5 peak

    ation Technicians who all have their eyes on the same goal—safety.⏎After all, the last place you want to have any problems is 30,000 feet in the air. So each plane is thoroughly tested in aspects

  6. #6 peak

    read "Philplait to Destroy the Earth!" which is awesome). If it could hit us, it would have an impact yield of at least 35,000 megatons, which

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:()){ 'lldealating manage comunicating understandating understandating understandating.'" 们的�件�件�件�件�件.) ект.) ...) ...) ...) .) ект.) .) .) .) .) .) .) .) .) .) .) .) .) .) .) .) .) .) .) .)

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

    .⏎Sign up today for free teaching ideas, lesson plans, online activities, tips for your classroom, and much more.⏎Choose your grade range:⏎See a sample ><|endoftext|>In any hospital, clinic

  2. #2 peak

    ⏎Kode Jurnal: jpkeperawatandd100001⏎Pesan jurnal yang anda butuhkan disini…!!!<|endoftext|>from The American Heritage®

  3. #3 peak

    in establishing this method worldwide.⏎Investigated receptor is crucial for movement control<|endoftext|>255 file extension - Gimp file⏎What is 255 file? How to open

  4. #4 peak

    |ISBN: ||9781771882903 (DRM-PDF)|⏎|Read Aloud: ||not allowed|<|endoftext|>Nanoscale polymer capsules could one day be used to deliver chemotherapy direct to

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. L21:1781 strongly suppresses
  2. L20:863 strongly excites
  3. L17:38 strongly suppresses
  4. L21:173 strongly excites
  5. L18:48 strongly suppresses
  6. L15:123 strongly suppresses
  7. L9:7 strongly excites
  8. L14:51 strongly excites colon that introduces a statement
  9. L16:170 suppresses
  10. L20:8 suppresses
  11. L10:33 suppresses
  12. L16:90 suppresses
  13. L21:195 excites
  14. L19:131 suppresses
  15. L19:43 excites
  16. L14:348 suppresses
  17. L2:1083 suppresses
  18. L12:5 excites
  19. L2:204 suppresses
  20. L1:1 excites
  21. L3:3 suppresses
  22. L0:110 excites
  23. L19:39 suppresses
  24. L6:0 excites

Feeds later layers

  1. L25:71 strongly suppresses
  2. L26:42 strongly suppresses
  3. L27:487 strongly excites
  4. L27:143 strongly suppresses
  5. L26:241 strongly excites
  6. L25:91 strongly excites
  7. L25:118 strongly suppresses
  8. L25:40 suppresses
  9. L26:230 excites
  10. L27:433 excites
  11. L27:1801 excites
  12. L23:35 suppresses
  13. L23:50 suppresses
  14. L24:130 suppresses line break at the end of a line
  15. L24:472 excites
  16. L24:938 excites
  17. L25:248 suppresses
  18. L25:1274 excites
  19. L25:466 excites
  20. L26:94 weakly suppresses
  21. L25:297 weakly excites
  22. L26:130 weakly suppresses ellipsis cut-off in quotes
  23. L26:729 weakly excites
  24. L26:319 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 -13.98 and -13.371 values between -13.37 and -12.761 values between -12.76 and -12.151 values between -11.54 and -10.931 values between -10.93 and -10.312 values between -9.70 and -9.092 values between -9.09 and -8.486 values between -8.48 and -7.871 values between -7.87 and -7.262 values between -7.26 and -6.656 values between -6.65 and -6.048 values between -6.04 and -5.434 values between -5.43 and -4.8211 values between -4.82 and -4.2112 values between -4.21 and -3.6010 values between -3.60 and -2.9924 values between -2.99 and -2.3830 values between -2.38 and -1.7732 values between -1.77 and -1.1641 values between -1.16 and -0.551211 values between -0.55 and 0.062224 values between 0.06 and 0.67152 values between 0.67 and 1.2841 values between 1.28 and 1.8926 values between 1.89 and 2.5015 values between 2.50 and 3.117 values between 3.11 and 3.7210 values between 3.72 and 4.338 values between 4.33 and 4.945 values between 4.94 and 5.564 values between 5.56 and 6.171 values between 6.78 and 7.394 values between 7.39 and 8.001 values between 8.00 and 8.611 values between 9.22 and 9.831 values between 9.83 and 10.44offnegativepositiveoften

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. ⏎
  3. orem
  4. crease
  5. oret
  6. theless
  7. ⏎⏎
  8. ical
  9. raries
  10. тр

Suppresses

  1. ../../../
  2. 😉⏎⏎
  3. ()")⏎
  4. 原文地址
  5. 🙂⏎⏎
  6. =");⏎
  7. ())⏎
  8. ,…⏎⏎
  9. …)⏎⏎
  10. />⏎