Neuron Explorer

Qwen3-0.6B

word-start 'm' pieces

A neuron in layer 26 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 a small 'm' that begins a word fragment, such as the 'm' in 'milder', 'mire' or 'mending'. Turning it up made the model repeat phrases or words like 'operating', which does not match the trigger.

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

    ándáil⏎Déantar do chuid sonraí íocaíochta a phróiseáil go slán. Ní dhéanann muid sonraí cártaí creidmheasa a stórá

  2. #2 peak

    packing them in oil or salt. Anchovy paste is also produced to use as an ingredient. Spanish boquerones are pickled in vinegar, which makes them milder.⏎Anchovies are notable in the culinary history of salt and<|endoftext|>M

  3. #3 peak

    1⏎Get out the microscope, because we’re going through this poem line-by-line.⏎No one has seen them made or heard them made,⏎But at spring mending-time we find them there.⏎- We get a bit distracted by all of

  4. #4 peak

    to return to normal⏎[weather, conditions] redevenir normal (e) → Even the weather has returned to normal.⏎(=regulation time) temps m réglementaire⏎at the end of normal time à l'issue du temps

  5. #5 peak

    vegetable stew, stuffed squash, green beans with rice, mouloukhiah, cauliflower stew.⏎4. Children meals preparation:⏎- Wash

  6. #6 peak

    will cluck like any hen,⏎And often to myself I think, why ye are just the sen!⏎– Singin’ blow, ye winds, in the mornin’,⏎– Blow, ye winds, high-ho!⏎– See

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: there is a t factor pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pean pe

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

    even look like other software, making them virtually undetectable until it's too late.⏎The easiest way to fend off malware is to invest in antivirus and anti-malware software, where an on-access real-time scanner hooks deep into your operating

  2. #2 peak

    of the witnesses by which the facts are to be proved, or if there be written evidence, an abstract of such evidence; 5. The personal character of the witnesses should be mentioned; whether the moral character is good or bad

  3. #3 peak

    They suffered physical and emotional abuse, and may have been trapped in generations of servitude.⏎However, the Bible also includes some stories of how servants influenced their owners/masters towards good deeds. (As one example, see 2 Kings 5

  4. #4 peak

    his or her own initiative can make the appointment. The literal meaning of the phrase “Guardian ad Litem” is guardian for the lawsuit, or guardian for the purposes of

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. L24:1190 excites
  2. L22:1138 suppresses
  3. L22:100 suppresses
  4. L17:230 excites
  5. L3:1101 suppresses
  6. L21:782 suppresses
  7. L18:553 excites
  8. L19:1531 excites
  9. L17:391 suppresses
  10. L17:241 suppresses
  11. L18:88 weakly excites
  12. L20:2288 weakly excites
  13. L19:381 weakly excites word-start 'acc' fragments
  14. L20:234 weakly suppresses
  15. L7:287 weakly suppresses
  16. L18:404 weakly excites 'get', 'go' and 'got' after a subject
  17. L20:212 weakly suppresses
  18. L7:342 weakly excites
  19. L19:334 weakly suppresses
  20. L8:2643 weakly excites
  21. L9:593 weakly suppresses
  22. L21:1679 weakly excites
  23. L1:1595 weakly suppresses
  24. L0:1010 weakly excites

Feeds later layers

  1. L27:1426 weakly excites
  2. L27:368 weakly excites
  3. L27:1319 weakly excites
  4. L27:1533 weakly suppresses
  5. L27:3070 weakly suppresses
  6. L27:2158 weakly suppresses
  7. L27:2241 weakly suppresses
  8. L27:2933 weakly suppresses the letter n opening a word
  9. L27:1097 weakly suppresses
  10. L27:2013 weakly excites
  11. L27:2282 weakly suppresses
  12. L27:1863 weakly suppresses
  13. L27:2810 weakly suppresses
  14. L27:1012 weakly excites
  15. L27:892 weakly excites
  16. L27:1976 weakly suppresses
  17. L27:939 weakly suppresses
  18. L27:480 weakly excites
  19. L27:1820 weakly suppresses
  20. L27:1826 weakly suppresses
  21. L27:2544 weakly excites
  22. L27:483 weakly excites
  23. L27:2142 weakly suppresses
  24. L27:2088 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 -18.11 and -17.021 values between -17.02 and -15.932 values between -13.75 and -12.661 values between -11.58 and -10.491 values between -10.49 and -9.401 values between -9.40 and -8.313 values between -8.31 and -7.226 values between -7.22 and -6.135 values between -6.13 and -5.044 values between -5.04 and -3.9512 values between -3.95 and -2.8620 values between -2.86 and -1.77139 values between -1.77 and -0.683114 values between -0.68 and 0.41524 values between 0.41 and 1.4919 values between 1.49 and 2.5816 values between 2.58 and 3.679 values between 3.67 and 4.7612 values between 4.76 and 5.852 values between 5.85 and 6.942 values between 6.94 and 8.032 values between 8.03 and 9.123 values between 9.12 and 10.212 values between 11.30 and 12.382 values between 12.38 and 13.471 values between 13.47 and 14.561 values between 16.74 and 17.831 values between 20.01 and 21.101 values between 21.10 and 22.191 values between 24.36 and 25.45offnegativepositiveoften

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. 指
  2. RCS
  3. rake
  4. ima
  5. Organ
  6. räume
  7. ajas
  8. operating
  9. emaker
  10. recruits

Suppresses

  1. ưỡng
  2. angep
  3. PyErr
  4. .squareup
  5. 网络传播
  6. 监事会
  7. olas
  8. .Pixel
  9. Pew
  10. tang