Neuron Explorer

Qwen3-0.6B

body and action words in instructions

A neuron in layer 13 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron tends to fire on words in step by step instructions, like 'take a deep [breath]', 'place one [finger]' or 'ask them [to] share'. It is weaker and mixed, and it turns off on words about events like 'launch' or 'start'. Turning it up did not give clear instructional text.

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

    you wonder how to subtract mixed numbers? Do you know how to handle unlike denominators or how to regroup or to borrow with mixed numbers? Take a deep breath and let us take it one step at a time. Click and begin.⏎G

  2. #2 peak

    Disagree Somewhat,” “Strongly Disagree”⏎Activities / Procedures:⏎1. WARM-UP/DO-NOW: Prior to class, write the words Security Check on the

  3. #3 peak

    .⏎Keeping a picture of a grandparent or a party or a vacation or an item or toy that they really want for their birthday or special holiday… (Big breath…) you get the picture… it helps take your child’s thoughts to a more

  4. #4 peak

    class to avoid confusion.⏎Partner Side Plank With Reach-Through Press⏎Partner 1 gets into side plank position: bottom arm extended beneath shoulder (or forearm supporting body), other arm extended upward and legs crossed

  5. #5 peak

    British response to the attack’. Within a couple of hours, these tweets had been shared thousands of times. By noon, a member of Parliament had read out the sign, which Prime Minister Theresa May called a “wonderful tribute”.⏎The

  6. #6 peak

    proper amount of fiber in the digestive tract makes it easier for the intestines to function properly. As an indication of how fiber works, try this: Place one finger from your left hand in the palm of your right hand and try to squeeze it

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 first character in the word "go" is the first character in the word "go" is a character that can be typed in a text editor: ' ' ' ' ` ` ` ` ` ` ` ` ` ` ` ` `

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

    of the Smithsonian.⏎Former first lady Laura Bush and Georgia Democratic Rep. John Lewis -- an icon of the civil rights era -- were among the other speakers celebrating the start of the museum's construction.⏎"I look forward to the day when I can am

  2. #2 peak

    are expected to be treated like a key go-to manual helping educators avoid risks when using social networks and ensure boundaries between their personal and professional lives.⏎Speaking at the launch, Parliamentary Secretary for Youths, Sports and Voluntary Organisations, Clifton

  3. #3 peak

    had to resume their studies and jobs in government institutions like schools, colleges, courts etc.<|endoftext|>I hope you are all well and that you enjoyed some of your weekend in the sunshine. It is now time to spend a brand new week completing our

  4. #4 peak

    It’s amazing to keep in touch with everybody and see some amazing work being produced.⏎We hope you have a lovely weekend and can’t wait to hear from you all today.⏎Mrs Beckwith, Miss

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. L10:1812 suppresses
  2. L10:2653 excites
  3. L12:1858 suppresses
  4. L11:1030 suppresses
  5. L12:1140 excites
  6. L12:1702 suppresses
  7. L11:2440 suppresses
  8. L7:1901 suppresses
  9. L9:1725 excites
  10. L10:2235 suppresses
  11. L12:1970 excites
  12. L11:271 suppresses
  13. L12:1634 excites
  14. L12:2060 weakly suppresses
  15. L12:122 weakly suppresses unclear: word before a clause continues
  16. L12:1431 weakly suppresses
  17. L10:716 weakly suppresses
  18. L12:2568 weakly suppresses
  19. L12:1883 weakly excites
  20. L9:936 weakly excites
  21. L11:1730 weakly suppresses
  22. L12:1639 weakly excites
  23. L11:1669 weakly excites
  24. L12:2007 weakly suppresses

Feeds later layers

  1. L16:2573 excites
  2. L23:925 excites
  3. L16:3027 suppresses
  4. L14:2686 excites
  5. L15:1829 excites
  6. L20:1666 excites
  7. L17:1571 suppresses
  8. L18:1226 suppresses
  9. L14:2206 excites
  10. L22:1542 excites
  11. L18:3025 suppresses
  12. L18:2260 excites
  13. L15:2213 suppresses
  14. L20:2902 suppresses
  15. L14:1819 excites
  16. L15:2741 excites
  17. L19:518 suppresses
  18. L19:2725 suppresses
  19. L14:753 suppresses
  20. L14:1515 suppresses
  21. L17:1865 excites
  22. L15:1599 suppresses
  23. L15:1715 suppresses
  24. L19:1348 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.

2 values between -2.88 and -2.771 values between -2.45 and -2.342 values between -2.13 and -2.021 values between -2.02 and -1.911 values between -1.80 and -1.692 values between -1.69 and -1.592 values between -1.59 and -1.481 values between -1.48 and -1.373 values between -1.37 and -1.263 values between -1.26 and -1.153 values between -1.15 and -1.055 values between -1.05 and -0.948 values between -0.94 and -0.8310 values between -0.83 and -0.7210 values between -0.72 and -0.6130 values between -0.61 and -0.5125 values between -0.51 and -0.4044 values between -0.40 and -0.29113 values between -0.29 and -0.18414 values between -0.18 and -0.071788 values between -0.07 and 0.04961 values between 0.04 and 0.14318 values between 0.14 and 0.2589 values between 0.25 and 0.3629 values between 0.36 and 0.4721 values between 0.47 and 0.589 values between 0.58 and 0.685 values between 0.68 and 0.793 values between 0.79 and 0.901 values between 0.90 and 1.011 values between 1.01 and 1.121 values between 1.12 and 1.221 values between 1.22 and 1.331 values between 1.33 and 1.44offnegativepositiveoften

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. 六年
  2. Years
  3. 七年
  4. years
  5. _formats
  6. economies
  7. _years
  8. וצא
  9. 年以上
  10. wars

Suppresses

  1. Congratulations
  2. 的通知
  3. gressor
  4. Congratulations
  5. 给您
  6. icket
  7. prise
  8. 的新
  9. 新加
  10. thanked