Neuron Explorer

Qwen3-0.6B

unclear: mixed multilingual, mid-word splits

A neuron in layer 24 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 wide mix of words that come right before a word is finished or continued, often in non-English text. Turning it up made the model drift into repeated, rambling sentences about keys and values.

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

    8 degrees Celsius, even for a few seconds, is enough to kill most bacteria, viruses, or parasites. Toxins produced by bacteria<|endoftext|>Dit artikel staat op icthealth.nl. Link: https://ictandhealth.com/ucla

  2. #2 peak

    άφει τετράγωνον σχῆμα τῷ λόφῳ in Einklang zu bringen, ist es unumgänglich nothwendig dass der Ausdruck Roma

  3. #3 peak

    ) equal in magnitude but have opposite direction⏎C) not equal in magnitude but have the same direction⏎D) not equal in magnitude but have opposite direction⏎E) None of the above⏎In a uniform circular motion, which of the following

  4. #4 peak

    ative Accusative mein (m)meinenmeinemein meine (pl.)meine All ein-words follow the same pattern: ein, kein, possessive adjectives⏎Personal Pronouns in

  5. #5 peak

    that buys offsets for planting trees (not as good) that might be cut down a few years later<|endoftext|>Hand-coloured lithograph entitled; 'Dampfmaschinen Pferd worauf man in einer stunde von Paris nach Petersburg

  6. #6 peak

    in 1934, and issued to units in 1936. It introduced an entirely new concept in automatic firepower – the ''Einheitsmaschineng

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: It is now 2023, and there is no longer no longer there is no longer is now again now now again now now now now now now now now now now now now now now now now now now now now now now

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

    ی منظر، دانشکده معماری و شهرسازی، دانشگاه علم و صنعت ایران، تهران، ایران.⏎عنوان مقاله [English]⏎In

  2. #2 peak

    replacing or augmenting the electronicssil...⏎Introduceți subiectul și categoria pentru o căutare cât mai relevantă.<|endoftext|>How to be

  3. #3 peak

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

  4. #4 peak

    -, etc.⏎Indefinite Articles in Accusative Case Nominative Accusative mein (m)meinenmeinemein meine (pl.)meine All ein-words follow the same pattern: ein, kein, possessive

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. L22:525 excites
  2. L23:253 suppresses
  3. L21:266 excites
  4. L18:2749 suppresses
  5. L21:627 suppresses
  6. L21:40 excites
  7. L22:539 suppresses
  8. L3:1757 weakly excites
  9. L20:377 weakly suppresses
  10. L4:2283 weakly excites
  11. L8:2028 weakly suppresses
  12. L20:267 weakly suppresses
  13. L18:560 weakly excites
  14. L19:1096 weakly suppresses
  15. L14:1661 weakly suppresses
  16. L18:2343 weakly suppresses
  17. L23:2394 weakly excites
  18. L15:65 weakly excites
  19. L7:22 weakly suppresses
  20. L21:795 weakly excites
  21. L22:589 weakly excites
  22. L0:1224 weakly suppresses
  23. L0:1301 weakly suppresses
  24. L10:2580 weakly suppresses

Feeds later layers

  1. L25:1942 weakly suppresses
  2. L27:840 weakly suppresses
  3. L27:1163 weakly suppresses
  4. L27:9 weakly excites
  5. L25:2002 weakly excites
  6. L25:23 weakly suppresses
  7. L25:2364 weakly excites
  8. L26:488 weakly excites
  9. L27:666 weakly suppresses
  10. L27:1357 weakly suppresses
  11. L27:1029 weakly excites
  12. L26:1591 weakly suppresses
  13. L25:2985 weakly excites
  14. L27:531 weakly excites
  15. L25:2243 weakly suppresses
  16. L25:1223 weakly excites
  17. L27:1229 weakly excites
  18. L25:2135 weakly suppresses
  19. L25:1597 weakly excites
  20. L27:1996 weakly suppresses
  21. L27:1826 weakly excites
  22. L26:2884 weakly suppresses
  23. L26:2356 weakly suppresses
  24. L25:57 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.

2 values between -13.14 and -12.431 values between -5.33 and -4.611 values between -4.61 and -3.902 values between -3.90 and -3.193 values between -2.48 and -1.7718 values between -1.77 and -1.06312 values between -1.06 and -0.353365 values between -0.35 and 0.36182 values between 0.36 and 1.079 values between 1.07 and 1.783 values between 1.78 and 2.494 values between 2.49 and 3.201 values between 3.20 and 3.912 values between 4.62 and 5.331 values between 5.33 and 6.041 values between 9.60 and 10.311 values between 14.57 and 15.28offnegativepositiveoften

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. (objects
  2. orian
  3. Ke
  4. Ke
  5. ince
  6. ippers
  7. enga
  8. states
  9. eneg
  10. ivant

Suppresses

  1. no
  2. no
  3. nos
  4. zp
  5. quires
  6. No
  7. zin
  8. będę
  9. →no
  10. No