Neuron Explorer

Qwen3-0.6B

unclear: word pieces inside names and words

A neuron in layer 26 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 reacts to a fragment in the middle of a longer name or word, like the 'of' in Sassoferrato or the 'Be' in Beethoven. Turning it up made the model's writing drift into repeated or off-topic text, so it doesn't clearly control anything we could name.

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

    and their visitors came from all walks of life and practiced asceticism with varying degrees of rigor. Certainly, an elite background did not preclude a monastic vocation

  2. #2 peak

    ò Perotti (1430-1480), secretary to Cardinal Bessarion and archbishop of Siponto, donated to Sassoferrato, the city of his birth, a collection of reliquaries,

  3. #3 peak

    Based on stylistic analysis, according to Martin Donnert, the mosaic icon at Sassoferrato was executed in the 14th century as a typical work of

  4. #4 peak

    symphony based on the life of the English military-imperial hero General Gordon, which, fortunately, did not get far.⏎Any composer approaching the symphony after Beethoven might well have had some trepidation about it. Brahms’ difficulties

  5. #5 peak

    Institute of Medical Sciences, Ansari Nagar, New Delhi 110029⏎|How to cite this article:|⏎Krishnan A, Behera P, Nongkyn

  6. #6 peak

    ’s nocturnes, Bach’s preludes and fugues, and the works of Clementi, Mozart and Beethoven. But Elsner also recognized Chopin’s unique talents and encouraged the

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: What if life on Earth reached immensively immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive immensive

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

    Soyuz 17, Salyut 6 EO-1, Salyut 7 EP-5. More...⏎Makarov Makarov, Oleg Grigoryevich (1933-20

  2. #2 peak

    the Rurik princes maintained their control over Kievan Rus and, later, Muscovy until 1598.⏎Rurik’s successor Oleg (d. 912) conquered Kiev (c. 8

  3. #3 peak

    pioneer and wealthy Chicago sportsman Logan “Jack” Vilas made the initial flight in order to demonstrate the viability of using aircraft in fire prevention. Wisconsin’s Chief Forester Edward Griffith hired him but Vilas refused pay, saying that he wanted only

  4. #4 peak

    was crowned as the first king of Jerusalem on Christmas Day, 1100, by Dagobert of Pisa, the Patriarch, who would have preferred to turn the

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. L2:172 suppresses
  2. L25:652 excites
  3. L3:459 weakly suppresses
  4. L15:1274 weakly excites
  5. L23:2441 weakly excites
  6. L22:268 weakly excites
  7. L10:370 weakly suppresses
  8. L20:256 weakly excites
  9. L6:1357 weakly excites
  10. L19:2045 weakly excites abbreviation pieces in organisation names
  11. L16:17 weakly excites
  12. L20:211 weakly excites
  13. L0:2987 weakly excites
  14. L22:1533 weakly suppresses
  15. L4:55 weakly excites
  16. L20:2371 weakly suppresses
  17. L15:1986 weakly excites
  18. L19:842 weakly suppresses
  19. L14:832 weakly suppresses modifier words before their head word
  20. L2:2387 weakly suppresses
  21. L7:1532 weakly suppresses
  22. L16:2096 weakly excites
  23. L23:2744 weakly excites
  24. L18:328 weakly suppresses

Feeds later layers

  1. L27:1663 excites
  2. L27:2671 weakly excites
  3. L27:2538 weakly excites
  4. L27:3047 weakly suppresses
  5. L27:1360 weakly excites
  6. L27:979 weakly excites
  7. L27:577 weakly excites
  8. L27:2059 weakly excites
  9. L27:903 weakly suppresses
  10. L27:2748 weakly suppresses
  11. L27:2788 weakly excites
  12. L27:496 weakly excites
  13. L27:1372 weakly excites
  14. L27:2079 weakly suppresses
  15. L27:2983 weakly excites
  16. L27:1441 weakly excites
  17. L27:1164 weakly excites
  18. L27:2763 weakly excites
  19. L27:2790 weakly suppresses
  20. L27:79 weakly excites
  21. L27:1461 weakly suppresses
  22. L27:1857 weakly excites
  23. L27:1385 weakly excites
  24. L27:3034 weakly suppresses

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 -23.63 and -22.681 values between -20.80 and -19.851 values between -19.85 and -18.911 values between -17.97 and -17.022 values between -14.19 and -13.251 values between -13.25 and -12.312 values between -11.36 and -10.421 values between -10.42 and -9.482 values between -9.48 and -8.533 values between -8.53 and -7.595 values between -7.59 and -6.652 values between -6.65 and -5.702 values between -5.70 and -4.769 values between -4.76 and -3.8112 values between -3.81 and -2.8715 values between -2.87 and -1.9339 values between -1.93 and -0.98999 values between -0.98 and -0.042702 values between -0.04 and 0.9091 values between 0.90 and 1.856 values between 1.85 and 2.797 values between 2.79 and 3.732 values between 3.73 and 4.681 values between 8.45 and 9.391 values between 13.17 and 14.11offnegativepositiveoften

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. str
  2. Str
  3. str
  4. _str
  5. Str
  6. →str
  7. -str
  8. STR
  9. .Str
  10. .str

Suppresses

  1. цу
  2. kapı
  3. gars
  4. \Modules
  5. /dir
  6. london
  7. lifting
  8. wood
  9. think
  10. Nz