Neuron Explorer

Qwen3-0.6B

unclear: word-internal fragments in foreign names

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 is weak and hard to pin down. It sometimes fires on pieces of foreign or Latin-script words and names split in the middle. Turning it up gave writing that did not clearly relate to this.

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

    (core) city ; Greek ἄστυ asty ) with the associated surrounding area ( χώρα chō

  2. #2 peak

    ойте скобки, употребляя глаголы в Present, Past, Future Indefinite, Present

  3. #3 peak

    perialisation. And this can only be done by reducing Russia’s territorial capabilities, says Professor Andrzej NOWAK in an interview with Mikołaj CZYŻ.⏎Mikołaj CZYŻ: Are the roots of Western sympathy for

  4. #4 peak

    ing, p. 293).⏎Grīwin Old Iranian languages. Grīw goes back to the Old Iranian grī

  5. #5 peak

    Old Ind. grīva- “neck,” grīvā- “the back part of the

  6. #6 peak

    is and then to discover how to avoid using it. As this is probably one of the most common mistakes made by English speaking students, it’s “imprescindible” that you get to grips with

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: I just thought of that. So, I need to get to some information from the internet to find out what is the problem here, and I need to get to some information about the problem. So, I need to get to some information

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

    the Catholic Church⏎The Council of Trent (Latin : Concilium Tridentinum), held between 1545 and 1563 in Trent

  2. #2 peak

    ἔλεγεν αὐτοῖς,⏎τὸ σάββατον διὰ τ

  3. #3 peak

    was the Libellus de abbreviaturis Hebraeorum, tam Talmudicorum quam Masoritarum et aliorum rabbinorum (Paris, 1561), later exploited by Guy *Le F

  4. #4 peak

    φος lexikographos, "lexicographer", from λεξικόν lexicon, neut. of λεξικός lexikos, "of or for words", from λέξι

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. L0:1224 suppresses
  2. L0:1301 suppresses
  3. L0:178 suppresses
  4. L0:508 weakly suppresses
  5. L0:641 weakly excites
  6. L0:263 weakly suppresses
  7. L0:79 weakly suppresses
  8. L0:3068 weakly suppresses
  9. L0:1 weakly suppresses
  10. L0:419 weakly suppresses
  11. L0:784 weakly excites 'of' after 'think' or 'out'
  12. L0:1214 weakly suppresses
  13. L0:510 weakly suppresses
  14. L0:1634 weakly suppresses
  15. L0:2408 weakly suppresses
  16. L0:472 weakly excites
  17. L0:535 weakly excites
  18. L0:2267 weakly excites
  19. L0:187 weakly suppresses
  20. L0:656 weakly suppresses 'to' after 'as opposed' or 'related'
  21. L0:1656 weakly excites
  22. L0:2478 weakly suppresses
  23. L0:2888 weakly suppresses
  24. L0:2146 weakly excites

Feeds later layers

  1. L4:513 suppresses
  2. L23:1678 weakly excites
  3. L10:2398 weakly excites
  4. L2:187 weakly suppresses
  5. L21:233 weakly suppresses
  6. L16:1793 weakly excites
  7. L26:1199 weakly excites
  8. L15:1214 weakly suppresses
  9. L2:1510 weakly suppresses
  10. L10:2964 weakly suppresses
  11. L8:550 weakly excites
  12. L13:3009 weakly excites
  13. L12:1108 weakly excites
  14. L16:1621 weakly suppresses
  15. L15:2792 weakly suppresses
  16. L22:2162 weakly excites
  17. L9:784 weakly excites
  18. L18:145 weakly excites
  19. L15:1051 weakly excites
  20. L10:636 weakly suppresses unclear: 'your' in contact boilerplate
  21. L2:778 weakly excites
  22. L3:66 weakly excites
  23. L14:282 weakly excites
  24. L18:2501 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 -1.92 and -1.861 values between -0.57 and -0.511 values between -0.51 and -0.443 values between -0.38 and -0.3225 values between -0.32 and -0.2659 values between -0.26 and -0.20162 values between -0.20 and -0.14461 values between -0.14 and -0.081133 values between -0.08 and -0.011285 values between -0.01 and 0.05517 values between 0.05 and 0.11177 values between 0.11 and 0.1758 values between 0.17 and 0.2314 values between 0.23 and 0.294 values between 0.29 and 0.366 values between 0.36 and 0.421 values between 0.48 and 0.54offnegativepositiveoften

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. erah
  2. leri
  3. 他们都
  4. quette
  5. jang
  6. シャル
  7. PLIER
  8. תי
  9. adalafil
  10. celona

Suppresses

  1. �
  2. VO
  3. vo
  4. ims
  5. �
  6. rows
  7. olith
  8. YT
  9. OF
  10. rots