Neuron Explorer

Qwen3-0.6B

digits inside historical years

A neuron in layer 22 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 one digit in the middle of a year written out in history text, such as the '6' in 1806 or the '9' in 1995. Turning it up gave answers full of numbers and counting problems, which is loosely related to 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

    1917. This is still the Constitution in force today.⏎President Benito Juarez Day celebrates his birthday of March 21, 1806. This holiday is observed on the third Monday of March. Benito Ju

  2. #2 peak

    0 meters in height, which feed a lava flow that crosses the road to the village of Portela .⏎The eruption of Fogo in 1995 - the formation of Pico Pequeno - via Geowiki⏎

  3. #3 peak

    eligible for School of Mathematics Doctoral Training Grant funding - please contact us for more information.<|endoftext|>História (São Paulo)⏎On-line version ISSN 1980-4369⏎DOMINGUES, Beatriz Helena.

  4. #4 peak

    light on the acts of gender-based violence they underwent during their detention, and the impact that it had in their lives after the recovery of democracy in 1983.⏎The Navy Mechanical School (Escuela de Mecánica de la

  5. #5 peak

    the pope's favor forced him to leave Rome and work for King Alfonso of Naples (1452-1455), but he resumed his position as papal secretary

  6. #6 peak

    President Luis Inácio da Silva, a man from the working class, became president in 2003, he called for a program to eliminate hunger completely. But that goal has

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: A person's 3i 100 100 100 100 100 100 100 100 100 100 1

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

    . Fixtures are designed to hold the workpiece in a specific position and orientation, allowing for consistent and accurate processing.⏎Deutsch: Frequenz / Español: Frecuencia / Português: Frequéncia / Français: Frêquence

  2. #2 peak

    . The period is the duration of one cycle in a repeating event, so the period is the reciprocal of the frequency.⏎Deutsch: Fabrik / Español: Fábrica / Português: Fábrica / Français: Us

  3. #3 peak

    687|⏎|Julkaisu||JOURNAL OF LANGUAGE TEACHING AND RESEARCH|⏎|DOI - pysyväislinkit|⏎|Tila||Julkaistu - 1 syysk. 20

  4. #4 peak

    an adjustable valve and a scaven<|endoftext|>Deutsch: Faktor / Español: Factor - Factor / Français: Facteur - Facteur / Italiano: Fattori - Fattori⏎A factor, a Latin word meaning may refer

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. L21:160 strongly suppresses
  2. L20:565 excites
  3. L20:550 excites
  4. L20:377 suppresses
  5. L21:437 suppresses
  6. L21:266 excites
  7. L18:178 excites
  8. L20:1198 suppresses
  9. L15:184 excites
  10. L18:2749 suppresses
  11. L21:627 excites
  12. L18:1703 excites
  13. L19:28 suppresses
  14. L19:622 suppresses
  15. L19:1096 suppresses
  16. L20:291 excites
  17. L18:550 suppresses
  18. L19:400 excites
  19. L21:1371 suppresses
  20. L10:39 suppresses
  21. L5:150 suppresses
  22. L17:98 weakly excites
  23. L21:2942 weakly suppresses comma before a clause that contrasts a claim
  24. L21:1873 weakly excites

Feeds later layers

  1. L26:824 suppresses
  2. L24:231 suppresses
  3. L25:362 suppresses
  4. L23:166 excites
  5. L24:2963 suppresses
  6. L26:1320 suppresses
  7. L24:202 excites
  8. L23:503 suppresses
  9. L26:2837 excites
  10. L25:1397 suppresses
  11. L25:1875 suppresses
  12. L23:2095 suppresses
  13. L26:1868 suppresses
  14. L26:1934 excites
  15. L26:2643 suppresses
  16. L23:352 suppresses
  17. L24:2200 suppresses
  18. L25:765 suppresses
  19. L26:1529 suppresses
  20. L26:1946 excites
  21. L25:2868 weakly excites
  22. L26:2584 weakly suppresses
  23. L27:1836 weakly excites
  24. L23:268 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 -16.97 and -16.231 values between -13.27 and -12.531 values between -7.35 and -6.611 values between -4.39 and -3.654 values between -3.65 and -2.912 values between -2.91 and -2.171 values between -2.17 and -1.434 values between -1.43 and -0.69174 values between -0.69 and 0.053373 values between 0.05 and 0.79315 values between 0.79 and 1.5328 values between 1.53 and 2.272 values between 2.27 and 3.011 values between 11.89 and 12.63offnegativepositiveoften

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. coh
  2. quot
  3. certain
  4. cons
  5. .annot
  6. umpt
  7. annot
  8. annes
  9. sch
  10. →de

Suppresses

  1. izada
  2. acias
  3. ificado
  4. ografia
  5. izado
  6. ificador
  7. uição
  8. ći
  9. ária
  10. atório