Neuron Explorer

Qwen3-0.6B

the digit zero in numbers

A neuron in layer 13 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 responds to the digit 0 inside numbers, especially in phone numbers like 1-800, in list numbers like 10, and in years like 1990s. It is strongest on that zero and weakest or negative on the zeros in some other numeric spots. Turning it up pushed the model toward unrelated number-heavy and odd text.

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

    electronic forms of online payment.⏎9. Be on guard against scams hiding behind online coupon offers.⏎10. Be careful if you download banking software onto a cell phone.⏎For details,

  2. #2 peak

    For further information and to get help call the Poison Help Line at 1-800-222-1222.⏎Synonyms and Keywords⏎wild

  3. #3 peak

    body language⏎6)Presence of mind⏎8)basic understanding of the nature and tone of the event.⏎9)A simple and dignified style.⏎10) Ability to deal with the technicians, event organizers, guest speakers.⏎Format of

  4. #4 peak

    careful and make sure to keep it away from pets and children.⏎In case of accidental ingestion, you can immediately call the Poison Center at 1-800-222-1222. They will provide assistance.⏎How

  5. #5 peak

    - 8. Buy second hand where possible and upcycle⏎- 9. Hang your clothes to dry rather than using a tumble dryer⏎- 10. Use all of your food leftovers⏎1. Reduce your meat intake⏎You

  6. #6 peak

    Disease induced changes in gene flow patterns among Tasmanian devil populations, Biological Conservation, 165 pp. 69-78. ISSN 0006-320

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: I will... Now I... I will... I will...... Let me know if you are, or I will, or, I, and I, I, I... I will... let me... I will,

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

    0s bought those semi-enclosed pedestal-style pay phones which, for most of us (Superman included) found lacking.⏎By the 1990s there were nearly three million pay phones in America. But now just a small

  2. #2 peak

    Mexico have migrated over 3,400 nautical miles to the west and over 2,300 nautical miles to the south.⏎Striped marlin are found throughout the

  3. #3 peak

    Night temperature's are often much lower around the high 60's and low 70's. A tortoise can maintain a higher body

  4. #4 peak

    used to run in Okayama-shi, part of western Japan, during 1910's-1930’s, Sanban Railroad Research Group commissioned to manufacture a station name board.⏎High

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. L8:41 weakly suppresses
  2. L7:1808 weakly excites
  3. L2:815 weakly excites
  4. L5:935 weakly suppresses
  5. L9:40 weakly suppresses
  6. L6:1322 weakly excites the last digit of decade years
  7. L6:1421 weakly suppresses closing bracket after an acronym
  8. L5:2798 weakly suppresses
  9. L4:549 weakly suppresses
  10. L7:3064 weakly excites
  11. L11:119 weakly suppresses
  12. L9:2785 weakly suppresses
  13. L0:581 weakly excites
  14. L6:1088 weakly suppresses
  15. L0:2855 weakly suppresses
  16. L12:2205 weakly suppresses
  17. L11:1909 weakly suppresses
  18. L2:1166 weakly excites
  19. L3:1181 weakly excites
  20. L2:261 weakly suppresses
  21. L2:1449 weakly suppresses
  22. L7:455 weakly suppresses
  23. L0:1811 weakly excites
  24. L4:291 weakly suppresses

Feeds later layers

  1. L16:2056 excites
  2. L24:1676 suppresses zeros in large round numbers
  3. L14:2970 suppresses
  4. L15:1012 suppresses
  5. L26:1603 weakly suppresses
  6. L15:2502 weakly suppresses the gap in number ranges
  7. L14:2527 weakly excites
  8. L25:1087 weakly suppresses word right before a number value
  9. L19:2678 weakly excites
  10. L22:1010 weakly suppresses
  11. L17:1422 weakly suppresses zeros in rounded thousands (e.g. 20,000)
  12. L20:1085 weakly excites
  13. L21:2853 weakly suppresses
  14. L19:1686 weakly excites
  15. L25:76 weakly excites
  16. L19:335 weakly excites
  17. L15:2196 weakly excites
  18. L18:2372 weakly excites
  19. L25:2967 weakly suppresses
  20. L19:2372 weakly excites
  21. L27:2223 weakly suppresses
  22. L22:1093 weakly excites
  23. L24:223 weakly suppresses
  24. L19:3002 weakly excites

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 -6.34 and -6.151 values between -5.59 and -5.401 values between -5.21 and -5.021 values between -5.02 and -4.832 values between -4.46 and -4.271 values between -4.08 and -3.891 values between -3.89 and -3.711 values between -3.52 and -3.331 values between -3.33 and -3.141 values between -3.14 and -2.962 values between -2.96 and -2.771 values between -2.77 and -2.582 values between -2.58 and -2.393 values between -2.39 and -2.202 values between -2.20 and -2.023 values between -2.02 and -1.833 values between -1.83 and -1.642 values between -1.64 and -1.452 values between -1.45 and -1.264 values between -1.26 and -1.086 values between -1.08 and -0.895 values between -0.89 and -0.7016 values between -0.70 and -0.51118 values between -0.51 and -0.33685 values between -0.33 and -0.141915 values between -0.14 and 0.05973 values between 0.05 and 0.24139 values between 0.24 and 0.4311 values between 0.43 and 0.613 values between 0.61 and 0.801 values between 0.80 and 0.991 values between 0.99 and 1.18offnegativepositiveoften

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. oph
  2. ائد
  3. .confirm
  4. ocaust
  5. _est
  6. ...⏎
  7. CO
  8. DU
  9. ços
  10. atin

Suppresses

  1. LEGAL
  2. чем
  3. 莳
  4. 把她
  5. 璜
  6. 质检
  7. localVar
  8. ла
  9. ➞
  10. Frameworks