Neuron Explorer

Qwen3-0.6B

split word pieces in names and years

A neuron in layer 9 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 fires on the middle pieces of longer names and words that get chopped into fragments, like the 'ople' in Constantinople or the 'ka' in Topeka. It also reacts to digits inside years. Turning it up did not produce a clear theme in the writing.

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

    were rivalry between the two brothers, and El Cid had supported Sancho. But Sancho died in 1072,

  2. #2 peak

    a separate peace with Germany and declared itself the enemy of all bourgeois states, France and Britain prominent among them.⏎What made you want to look up "Constantinople Agreement"? Please share what surprised you most...<|endoftext|>Your healthcare provider will ask about

  3. #3 peak

    Aid Company; School buildings; Schools; Shawnee County, Kansas Territory; Topeka, Kansas Territory⏎Letter, E. [Ephraim] Nute

  4. #4 peak

    President Boris Yeltsin. Like the Americans, the Russians use a mandatory 10 minute launch window (the time needed for a submarine-la<|endoftext|>Constantinople AgreementArticle Free Pass⏎Constantinople Agreement, (March 18,

  5. #5 peak

    with the example of the Blessed Virgin, St. Joseph, the Shepherds, and, who could forget the Adoration of the Magi. The popular devotion as we know it has been particularly strong in Spain since the

  6. #6 peak

    rix of the parabola. [Adopted from iCoachMath]⏎Focus:Focus of a parabola lies on the axis of symmetry.⏎Directrix:Directrix is a line that is perpendicular to the axis of symmetry of a

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 been told that there are 1000000000000000000000000000000000000000

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

    id. Source: Twitter⏎1929 motor coach no.1 at Palma station. Source: Twitter⏎Sóller station in 2019, with the Puerto tramlines on the left. Source: Twitter⏎The

  2. #2 peak

    general and first president, respectively.⏎These two 1847 stamps hold their value through market fluctuations and through thick times and thin.⏎The 2020 Scott Specialized Catalogue of United States

  3. #3 peak

    in equivalent purchasing power in 2010). So the picture has never been quite so pretty as the image that is so often laid out for us when we are told of the halcyon past in which

  4. #4 peak

    until its demolition in the mid-1960s. The oval dining room was removed from the mansion and recreated in the Baltimore Museum of Art where it remains a part of the American Decorative Arts wing.⏎This demolition sparked a renewed awareness

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. L8:764 excites
  2. L7:448 excites
  3. L7:305 excites
  4. L8:418 excites
  5. L8:474 excites
  6. L6:2881 suppresses
  7. L8:800 weakly excites
  8. L6:113 weakly suppresses
  9. L6:1717 weakly suppresses
  10. L8:2096 weakly excites
  11. L6:1667 weakly suppresses
  12. L5:276 weakly suppresses
  13. L2:2692 weakly suppresses
  14. L8:2719 weakly suppresses
  15. L8:3069 weakly excites
  16. L7:122 weakly suppresses
  17. L5:158 weakly excites
  18. L8:1432 weakly excites
  19. L3:139 weakly excites
  20. L6:2356 weakly excites
  21. L3:238 weakly suppresses
  22. L3:596 weakly excites
  23. L1:1464 weakly suppresses
  24. L7:113 weakly suppresses

Feeds later layers

  1. L10:375 suppresses
  2. L14:403 suppresses
  3. L19:1026 suppresses
  4. L13:1926 excites
  5. L10:2351 excites science news body text
  6. L13:2480 suppresses
  7. L11:464 suppresses
  8. L10:3044 excites
  9. L10:429 suppresses
  10. L13:2491 weakly excites
  11. L12:1970 weakly suppresses
  12. L12:1713 weakly suppresses
  13. L12:61 weakly suppresses
  14. L16:546 weakly suppresses
  15. L20:2388 weakly suppresses
  16. L11:2896 weakly excites
  17. L23:1047 weakly excites
  18. L10:1517 weakly suppresses
  19. L10:2196 weakly suppresses
  20. L11:2736 weakly excites
  21. L10:716 weakly excites
  22. L21:2819 weakly excites
  23. L11:1636 weakly suppresses
  24. L17:254 weakly suppresses unclear: word before a twist or contrast

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 -4.02 and -3.871 values between -3.71 and -3.551 values between -3.24 and -3.082 values between -2.60 and -2.453 values between -2.45 and -2.294 values between -2.29 and -2.1310 values between -2.13 and -1.976 values between -1.97 and -1.8211 values between -1.82 and -1.6629 values between -1.66 and -1.5038 values between -1.50 and -1.3455 values between -1.34 and -1.1958 values between -1.19 and -1.0378 values between -1.03 and -0.87138 values between -0.87 and -0.71190 values between -0.71 and -0.56258 values between -0.56 and -0.40365 values between -0.40 and -0.24486 values between -0.24 and -0.08714 values between -0.08 and 0.08562 values between 0.08 and 0.23376 values between 0.23 and 0.39207 values between 0.39 and 0.55129 values between 0.55 and 0.7177 values between 0.71 and 0.8645 values between 0.86 and 1.0225 values between 1.02 and 1.1814 values between 1.18 and 1.3418 values between 1.34 and 1.501 values between 1.50 and 1.652 values between 1.65 and 1.811 values between 1.81 and 1.972 values between 1.97 and 2.131 values between 2.13 and 2.28offnegativepositiveoften

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. ewidth
  2. еств
  3. BaseActivity
  4. leness
  5. (show
  6. 玒
  7. NCY
  8. vens
  9. красот
  10. zach

Suppresses

  1. unks
  2. -⏎⏎
  3. 升
  4. Kn
  5. ...⏎⏎
  6. reet
  7. 割
  8. kn
  9. -⏎⏎
  10. 吸