Neuron Explorer

Qwen3-0.6B

unclear: word fragments in foreign text

A neuron in layer 25 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 pieces of longer words in non-English text, such as the middle of a Swedish or Persian word, and it is hard to say what they have in common. Turning it up made the model repeat itself or drift into Chinese characters, so it does not show a clear job.

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

    to increase, which is a growing public health concern.⏎|Status||Published - 2014 jan. 1|⏎- Folkhälsovetenskap, global hälsa, socialmedicin och epidemiologi<|endoftext|>Russia can

  2. #2 peak

    : Madison McDermott⏎The Green New Deal Federal Resolution was introduced in 2019 by Rep. Ocasio-Cortez and Sen. Ed Markey as a plan to tackle a trifecta of crises: the climate

  3. #3 peak

    .⏎2 پژوهشگر دوره تخصصی دکتری معماری منظر، دانشکده معماری و شهرسازی، دانش

  4. #4 peak

    a representation of both the classes whereas TEU consists of officers only. Though<|endoftext|>نوع مقاله : مقاله پژوهشی⏎1 دانشیار معماری، دانشکده معماری

  5. #5 peak

    Pine Ridge, he advocated for peace with the European Americans⏎Date de dernière mise à jour : 05/07/2021⏎Ajouter un commentaire<|endoftext|>By DANIEL SUDDEATH⏎NEW ALBANY —

  6. #6 peak

    uffassung des Dionys. I 88: περιγράφει τετράγωνον σχῆμα τῷ λόφῳ in Einklang zu bringen

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:这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这这

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

    and 2014, World Bank says. Wealth remains unequal, and some countries are falling behind.⏎El Banco Mundial analiza la riqueza de los países, tomando en cuenta no solo el PIB, sino también el

  2. #2 peak

    research on BPA. Glass bottles, however, also leach chemicals, including aluminum, explains a study by Clemens Reimann of the Norges Geologiske Undersøkelse in Norway and colleagues. Aluminum is sometimes added to

  3. #3 peak

    ) право голосовать⏎2. Прочитайте текст Mayor-Council и заполните пропуски текста словами и выражениями, данными в

  4. #4 peak

    ⏎1. Соотнесите значения слов и выражений, данных в левой колонке, с их английскими эквивалентами, данными в правой колонке⏎1) legislative branch a) Член

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. L21:1248 suppresses
  2. L20:1835 suppresses
  3. L23:618 excites
  4. L21:1376 suppresses
  5. L19:1424 excites
  6. L21:1698 excites
  7. L24:434 suppresses
  8. L24:2963 excites
  9. L19:2842 suppresses
  10. L19:639 excites
  11. L23:2052 suppresses
  12. L22:735 suppresses
  13. L18:2872 excites
  14. L17:1371 excites
  15. L23:66 excites
  16. L17:1462 excites
  17. L20:3022 excites
  18. L20:541 excites
  19. L22:186 excites
  20. L20:54 excites
  21. L19:400 excites
  22. L20:1842 weakly suppresses
  23. L21:607 weakly excites
  24. L15:3007 weakly excites

Feeds later layers

  1. L26:507 weakly excites
  2. L26:405 weakly suppresses
  3. L27:1458 weakly excites
  4. L27:694 weakly excites
  5. L27:127 weakly suppresses
  6. L26:1052 weakly excites
  7. L26:479 weakly suppresses
  8. L27:1771 weakly excites
  9. L27:472 weakly suppresses
  10. L26:763 weakly excites
  11. L26:968 weakly excites
  12. L27:399 weakly excites
  13. L26:1356 weakly suppresses
  14. L26:1515 weakly excites
  15. L26:2417 weakly excites
  16. L27:994 weakly suppresses
  17. L26:1647 weakly excites
  18. L27:273 weakly suppresses
  19. L26:716 weakly suppresses
  20. L27:1168 weakly excites
  21. L27:2608 weakly suppresses
  22. L27:2317 weakly excites
  23. L27:1280 weakly excites
  24. L26:1913 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 -3.60 and -3.461 values between -1.68 and -1.531 values between -1.53 and -1.385 values between -1.38 and -1.245 values between -1.24 and -1.0913 values between -1.09 and -0.9419 values between -0.94 and -0.7941 values between -0.79 and -0.64128 values between -0.64 and -0.50263 values between -0.50 and -0.35596 values between -0.35 and -0.201315 values between -0.20 and -0.051115 values between -0.05 and 0.10240 values between 0.10 and 0.2486 values between 0.24 and 0.3948 values between 0.39 and 0.5418 values between 0.54 and 0.698 values between 0.69 and 0.843 values between 0.84 and 0.982 values between 2.17 and 2.32offnegativepositiveoften

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. 那些
  2. 这些
  3. degli
  4. 彼ら
  5. les
  6. 三个
  7. 两类
  8. 各
  9. Les
  10. 們

Suppresses

  1. 这
  2. 這
  3. 而这
  4. 这也
  5. 但这
  6. �
  7. isplay
  8. _deinit
  9. hers
  10. .hl