Neuron Explorer

Qwen3-0.6B

unclear: mid-phrase words across many languages

A neuron in layer 8 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 a random assortment of words in the middle of a phrase, in many languages, with no clear shared meaning. Turning it up or down changed the text the model wrote in unrelated ways, often switching languages.

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

    turning off lights).⏎We appreciate the efforts of the contributors to this special issue who are addressing some of<|endoftext|>L’Encyclopédie de l’histoire du Québec / The Quebec History Encyclopedia⏎Ethics and Morals of the Indians of

  2. #2 peak

    88: περιγράφει τετράγωνον σχῆμα τῷ λόφῳ in Einklang zu bringen, ist es unumgänglich no

  3. #3 peak

    en and Plan du Havre de Noel, dans la Terre de Kerguelen, as well as an inset at bottom left showing two

  4. #4 peak

    de Doopsgezinde Gemeente to Amsterdam, 2 vols. Amsterdam: Uitgegeven en ten geschenke aangeboden door den Kerkeraad dier Gemeente, 1883-

  5. #5 peak

    Is.⏎|Prepared by:||Gordon Rudloff⏎Ohio Department of Agriculture<|endoftext|>Logic & ProofsOpen Learning Initiative⏎Cosa impari in questo corso?⏎Logic & Pro

  6. #6 peak

    these prepositions must be in accusative case: durch-throughfuer-for gegen-againstohne-without um-around Er kauft das Geschenk fuer mich.<|endoftext|>In case you are searching for easy Preschool Activities, you can

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: vous pouvez-tu dire que vous ne comprenez pas la philosophie des systèmes de gestion de ressources de données, c'est un concept de thèse, mais c'est un système de notation? Pour les réflexions

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

    O snap/ring gauges (for checking shafts) and plug gauges (for

  2. #2 peak

    der Zijpp|⏎<|endoftext|>The Great Pacific Garbage Patch may qualify as the largest garbage

  3. #3 peak

    04)⏎(4) « La compression veineuse est le traitement de base des affections ve<|endoftext|>With concern over

  4. #4 peak

    Antalya-Türkiye’de doğdu. Lisans öğrenimini MSÜ Deniz Harp Okulunda tamamlamıştır.⏎

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. L3:6 strongly suppresses
  2. L5:136 strongly suppresses
  3. L3:1090 strongly suppresses
  4. L5:150 suppresses
  5. L2:166 suppresses
  6. L2:2022 suppresses
  7. L3:113 excites
  8. L2:24 excites
  9. L3:530 excites
  10. L7:130 excites
  11. L1:1227 excites
  12. L6:479 suppresses
  13. L1:651 suppresses
  14. L4:1147 suppresses
  15. L2:1733 suppresses
  16. L6:331 weakly excites
  17. L0:1090 weakly excites
  18. L4:1272 weakly excites
  19. L3:318 weakly excites
  20. L4:1789 weakly excites
  21. L2:1934 weakly excites
  22. L2:1700 weakly suppresses
  23. L7:92 weakly excites
  24. L0:83 weakly excites

Feeds later layers

  1. L15:65 strongly suppresses
  2. L16:477 strongly suppresses
  3. L17:487 strongly suppresses
  4. L17:649 strongly suppresses
  5. L10:39 strongly suppresses
  6. L14:381 strongly suppresses
  7. L17:1021 strongly suppresses
  8. L11:137 strongly suppresses
  9. L13:1204 strongly suppresses
  10. L9:453 strongly excites
  11. L11:416 strongly suppresses
  12. L13:929 strongly suppresses
  13. L20:377 strongly suppresses
  14. L21:266 strongly suppresses
  15. L16:1347 suppresses
  16. L18:560 excites
  17. L12:1423 suppresses
  18. L14:1424 suppresses
  19. L20:267 suppresses
  20. L16:979 suppresses
  21. L18:566 suppresses
  22. L18:72 excites
  23. L17:2506 suppresses
  24. L18:567 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 -2.83 and -2.741 values between -2.55 and -2.461 values between -1.27 and -1.181 values between -1.18 and -1.082 values between -1.08 and -0.991 values between -0.90 and -0.812 values between -0.72 and -0.632 values between -0.63 and -0.531 values between -0.44 and -0.353 values between -0.35 and -0.263 values between -0.17 and -0.0743 values between -0.07 and 0.02202 values between 0.02 and 0.11803 values between 0.11 and 0.201061 values between 0.20 and 0.29841 values between 0.29 and 0.39534 values between 0.39 and 0.48254 values between 0.48 and 0.57117 values between 0.57 and 0.6627 values between 0.66 and 0.758 values between 0.75 and 0.84offnegativepositiveoften

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. hers
  2. 的独特
  3. 之势
  4. rove
  5. 了一口
  6. ropes
  7. roach
  8. ToBounds
  9. ировки
  10. 扒

Suppresses

  1. ác
  2. dé
  3. STRU
  4. enny
  5. .opend
  6. annel
  7. ح
  8. oire
  9. .ReadLine
  10. "));⏎