Neuron Explorer

Qwen3-0.6B

chinese characters mid-phrase

A neuron in layer 27 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 Chinese characters, especially inside Chinese phrases and poems, and it is quiet on English. Turning it up mostly produced repeated English words like 'lotto' and 'lotus', so the steering effect does not show the Chinese theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens an email to the maker with the neuron already filled in).

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

    對 CNY 感興趣?⏎Early Currency in China⏎With a history of over 300

  2. #2 peak

    also means that pollutants that originate indoors are getting trapped inside.⏎“People normally<|endoftext|>Spring Dawn by Meng Haoran 春眠不觉晓 处处闻啼茑 夜来风雨声 花落知多少⏎Pin Yin Version

  3. #3 peak

    to be an accessible topic for middle school students.19<|endoftext|>毋忘在莒(Pinyin: Wu Wang Zai Ju) is a famous Chinese

  4. #4 peak

    human behaviour. Instead of the legal system, early Confucian scholars emphasized the concepts of morality and ritualism. The term “ritual propriety” (禮, pinyin: lǐ), describes the “proper” social relationships and the

  5. #5 peak

    Haoran Major poet during the Tang Dynasty Setting of Spring Dawn⏎Meaning for Line 1 春 : spring 眠 : sleep 不 : not 觉 : to feel, wake up 晓 : dawn⏎Meaning for L

  6. #6 peak

    221BC). Unlike most Chinese idioms, 毋忘在莒 strangely has actually two stories of origin, which gave this idiom

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: Abstract Designs Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract Designs Abstract

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

    lunar Vedic calendar. Holidays are commemorated each year based on the lunar month and the phase of the Moon, according to Vedic astrology, or Jyotish

  2. #2 peak

    same BP reading may be extremely normal.⏎• For healthy people, especially athletes, low systolic pressure may<|endoftext|>The Great Trek (Afrikaans: Die Groot Trek; Dutch: De Grote Trek) was an eastward and north-east

  3. #3 peak

    ereed Article]⏎Infectious diseases of wildlife reduce population size and may erode genetic diversity, constituting an extinction threat. The Tasmanian devil (Sarcophilus harrisii) is threatened with extinction by an infectious cancer,

  4. #4 peak

    and eastern⏎Botswana, north-eastern Namibia and South Africa, specifically in the Kruger⏎National Park, KwaZulu-Natal and the Eastern Cape. It prefers woodland dominated⏎by

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. L0:178 suppresses
  3. L15:184 excites
  4. L0:1301 suppresses
  5. L7:92 excites
  6. L18:251 excites
  7. L20:270 excites
  8. L13:44 suppresses inside web addresses and citations
  9. L23:126 suppresses
  10. L17:473 suppresses
  11. L22:349 excites
  12. L10:155 excites
  13. L2:483 suppresses
  14. L19:400 excites
  15. L15:390 excites
  16. L18:66 suppresses
  17. L20:541 excites
  18. L18:178 excites
  19. L9:74 excites
  20. L18:125 suppresses
  21. L19:622 suppresses
  22. L6:13 excites
  23. L23:66 excites
  24. L1:25 weakly excites

Feeds later layers

None: this is the last layer.

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 -8.67 and -8.001 values between -5.31 and -4.641 values between -4.64 and -3.972 values between -3.30 and -2.634 values between -2.63 and -1.952 values between -1.95 and -1.2861 values between -1.28 and -0.613342 values between -0.61 and 0.06460 values between 0.06 and 0.7313 values between 0.73 and 1.415 values between 1.41 and 2.084 values between 2.08 and 2.753 values between 2.75 and 3.422 values between 4.09 and 4.771 values between 4.77 and 5.441 values between 5.44 and 6.111 values between 6.11 and 6.781 values between 6.78 and 7.452 values between 9.47 and 10.141 values between 17.53 and 18.20offnegativepositiveoften

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. crossed
  2. panels
  3. drops
  4. toPromise
  5. lifted
  6. marvin
  7. layers
  8. readonly
  9. dro
  10. sz

Suppresses

  1. 款
  2. iences
  3. 岗
  4. 规
  5. 话说
  6. okies
  7. 事
  8. uest
  9. �性
  10. 想不到