Neuron Explorer

Qwen3-0.6B

the digit 9 inside numbers

A neuron in layer 19 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 digit 9 when it appears inside a number, such as a percentage, a phone number like 9-1-1, or a rate like 98%. Its negative side fires on the word email and its pieces, and turning it down pushed the model toward writing about email.

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

    ⏎To use the formula, first figure out what you want your error of tolerance to be. For example, you may be happy with a confidence level of 95 percent (giving a margin error of 0.05), or

  2. #2 peak

    Overall, these tests have excellent detection rates (~99%) with very low false-positive rates (~0.2%). In other words, these tests are about 99.0% sensitive and 99.8% specific.⏎With performance

  3. #3 peak

    or discomfort, or a severe cough, contact your health care provider, walk-in clinic, or emergency department. If you are having a medical emergency, call 9-1-1.⏎Tips to reduce your smoke exposure⏎Smoke levels may be

  4. #4 peak

    world. All steels, including structural steels, are completely recyclable materials. Steel material has the highest rate in the market with a recycling rate of 98%. It always returns to the steelmaking facilities as scrap steel and it can

  5. #5 peak

    first-come, first-serve basis.⏎For information call 9<|endoftext|>The Parliamentary Commissioner for the Environment has announced:⏎Parliamentary Commissioner for the

  6. #6 peak

    spread through your bloodstream and cause life-threatening damage to your heart or kidneys. It can also cause nerve damage that leads to paralysis.⏎Call your local emergency number (911 in the US) if:⏎- You have severe trouble breathing.⏎Call

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 have a 4-bit address address. I have a 4-bit address address. I have a 4-bit address address. I have a 4-bit address address. I have a 4-bit address address. I have a

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

    below. You may download and print the language guide for personal use.⏎Please note: this is not a static document and will evolve over time. Please email [email protected] or use our feedback form if there are other words you think

  2. #2 peak

    -Sheva, Israel⏎- *Corresponding Author:⏎- Yeongha Wook⏎Department of Pediatrics,⏎Soroka University Medical Center,⏎E-mail: [email protected]⏎Received: 01-Nov-2022

  3. #3 peak

    these relate to what we know today.⏎In association with STFC Rutherford Appleton Laboratory⏎If your child has additional needs, please get in touch on [email protected] to discuss how we can best support you.⏎For information on our booking

  4. #4 peak

    to do it too!⏎Winter Olympics | Pocantico Hills School⏎This page was last edited December 02, 2005.⏎e-mail us at:<|endoftext|>Courtesy ESA/Rosetta/MPS for OSIRIS Team

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. L5:2675 suppresses
  2. L9:2112 excites
  3. L7:2039 suppresses
  4. L6:2602 suppresses
  5. L18:1104 suppresses
  6. L9:2017 suppresses
  7. L5:2534 excites
  8. L12:2751 suppresses
  9. L8:2782 suppresses
  10. L6:2013 excites
  11. L18:3020 weakly suppresses
  12. L8:1408 weakly suppresses
  13. L11:626 weakly excites
  14. L5:2409 weakly excites
  15. L6:2428 weakly excites
  16. L2:2099 weakly excites
  17. L11:1502 weakly excites
  18. L7:2461 weakly suppresses
  19. L15:2809 weakly suppresses
  20. L7:1542 weakly suppresses
  21. L6:1722 weakly suppresses
  22. L9:1772 weakly excites
  23. L5:3034 weakly excites
  24. L7:1430 weakly suppresses

Feeds later layers

  1. L23:1232 suppresses
  2. L23:3053 suppresses
  3. L27:2421 suppresses
  4. L22:2890 suppresses
  5. L24:1698 suppresses
  6. L27:1952 weakly suppresses
  7. L22:2622 weakly suppresses
  8. L21:2649 weakly excites
  9. L25:2586 weakly excites
  10. L26:1315 weakly suppresses
  11. L23:1885 weakly excites
  12. L20:454 weakly excites
  13. L20:2138 weakly excites
  14. L21:435 weakly suppresses
  15. L25:2588 weakly suppresses
  16. L25:1874 weakly suppresses
  17. L21:2847 weakly suppresses
  18. L21:1625 weakly excites
  19. L21:2186 weakly excites
  20. L23:1307 weakly suppresses
  21. L23:2793 weakly excites
  22. L21:2579 weakly suppresses
  23. L22:2860 weakly excites
  24. L24:3062 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 -36.75 and -35.381 values between -23.05 and -21.681 values between -18.94 and -17.571 values between -16.20 and -14.831 values between -10.72 and -9.351 values between -9.35 and -7.983 values between -5.24 and -3.875 values between -3.87 and -2.5018 values between -2.50 and -1.132721 values between -1.13 and 0.241151 values between 0.24 and 1.612 values between 1.61 and 2.981 values between 9.83 and 11.201 values between 16.68 and 18.05offnegativepositiveoften

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. 鳌
  2. erah
  3. vre
  4. рез
  5. roaring
  6. ре
  7. mpeg
  8. otics
  9. Nos
  10. ledon

Suppresses

  1. 邮件
  2. (email
  3. =email
  4. emails
  5. email
  6. Mail
  7. 邮箱
  8. getEmail
  9. 电子邮件
  10. Gmail