Neuron Explorer

Qwen3-0.6B

word fragments inside non-English words

A neuron in layer 1 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 short piece in the middle of a word from another language, such as 'nos' in a Latvian word or 'als' in a Polish name. Its values are small and mixed, and turning it up did not give a clear pattern in the writing.

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

    products. In the spring of 1934, the Franks r…⏎E-pasta adrese, uz kuru nosūtīt darba saiti:⏎Saite uz darbu:<|endoftext|>

  2. #2 peak

    is the most obvious similarity between the two religions. In fact a Muslim is defined as anyone who says "Th…⏎E-pasta adrese, uz kuru nosūtī

  3. #3 peak

    ons of the Prussian King Friedrich Wilhelm I, who finally expelled them from the district. The [[Amsterdam Mennonite Library (Bibliotheek en Archief van de Vereenigde Doopsgezinde Gemeente te

  4. #4 peak

    inct in the Wild until 2008, but since a series of reintroductions, have developed free ranging populations in the wild. All Przewalski’s horses alive today are descendent of 14 individuals from a captive

  5. #5 peak

    established to guide players, and the people that coach them, through a level-appropriate system of basketball development. Using scientific guiding principles developed by coach educators Istvan Balyi and Richard Way, and found in their book Long-Term Athlete

  6. #6 peak

    na in the wilderness (John 6:49-51).⏎Dragons still

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: this is a new question in the field of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of the study of

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

    ментарии при наличии в них нецензурной брани, оскорблений или спама<|endoftext|>Conflict and Solidarity in Mudejar

  2. #2 peak

    ЛЬНАЯ РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎КО

  3. #3 peak

    simple problem, one that a fifth grader may find in a math book. What about the problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 13.09

  4. #4 peak

    under the czar’s vicegerent in Warsaw. Rzewuski edited the government newspaper Dziennik warszawski.⏎WORKSDziela, vols. 1–6. Warsaw, 1877–8

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. L0:178 suppresses
  2. L0:1224 suppresses
  3. L0:1301 suppresses
  4. L0:1096 weakly suppresses
  5. L0:894 weakly excites
  6. L0:2942 weakly excites
  7. L0:1 weakly suppresses
  8. L0:91 weakly suppresses
  9. L0:794 weakly excites
  10. L0:3070 weakly suppresses
  11. L0:104 weakly suppresses
  12. L0:2609 weakly excites
  13. L0:2569 weakly excites
  14. L0:25 weakly excites
  15. L0:386 weakly suppresses
  16. L0:2415 weakly suppresses
  17. L0:188 weakly excites
  18. L0:2642 weakly excites
  19. L0:356 weakly suppresses
  20. L0:540 weakly suppresses
  21. L0:1315 weakly excites
  22. L0:63 weakly excites
  23. L0:1539 weakly excites
  24. L0:1936 weakly suppresses

Feeds later layers

  1. L2:87 excites
  2. L3:809 suppresses
  3. L5:763 suppresses
  4. L20:1245 excites
  5. L17:385 weakly suppresses
  6. L26:524 weakly excites
  7. L22:948 weakly excites
  8. L6:2567 weakly excites "by" after a passive verb
  9. L10:2023 weakly suppresses
  10. L18:1135 weakly suppresses
  11. L14:1353 weakly suppresses
  12. L17:303 weakly suppresses
  13. L13:1497 weakly excites
  14. L7:132 weakly suppresses
  15. L24:2014 weakly excites
  16. L19:1278 weakly excites
  17. L7:572 weakly suppresses
  18. L3:232 weakly suppresses 'on' after a noun like 'impact on'
  19. L15:2557 weakly excites
  20. L20:2014 weakly suppresses
  21. L16:1727 weakly excites
  22. L5:1861 weakly excites
  23. L22:2753 weakly suppresses
  24. L23:1586 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 -0.44 and -0.421 values between -0.38 and -0.361 values between -0.36 and -0.343 values between -0.32 and -0.304 values between -0.30 and -0.2810 values between -0.28 and -0.2612 values between -0.26 and -0.2423 values between -0.24 and -0.2124 values between -0.21 and -0.1920 values between -0.19 and -0.1744 values between -0.17 and -0.1563 values between -0.15 and -0.1373 values between -0.13 and -0.11116 values between -0.11 and -0.09154 values between -0.09 and -0.07210 values between -0.07 and -0.05376 values between -0.05 and -0.03772 values between -0.03 and -0.01926 values between -0.01 and 0.01458 values between 0.01 and 0.03246 values between 0.03 and 0.05135 values between 0.05 and 0.0787 values between 0.07 and 0.0938 values between 0.09 and 0.1141 values between 0.11 and 0.1416 values between 0.14 and 0.1614 values between 0.16 and 0.186 values between 0.18 and 0.206 values between 0.20 and 0.228 values between 0.22 and 0.248 values between 0.24 and 0.264 values between 0.26 and 0.283 values between 0.28 and 0.302 values between 0.30 and 0.321 values between 0.34 and 0.362 values between 0.36 and 0.38offnegativepositiveoften

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. groom
  2. 不多
  3. phấn
  4. poly
  5. crawler
  6. unger
  7. atology
  8. ivery
  9. тся
  10. vation

Suppresses

  1. uit
  2. 日前
  3. recent
  4. eking
  5. est
  6. 日正式
  7. host
  8. behalf
  9. host
  10. heck