Neuron Explorer

Qwen3-0.6B

'even' in 'and even' lists

A neuron in layer 7 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word 'even' when it adds a surprising last item to a list, as in 'and even death'. Turning it up did not make the model's text show that pattern, so the link to the output is unclear.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. 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

    sun was converted to energy in a fraction of a second.⏎The discovery has been accepted for publication in Physical Review Letters.⏎Gravitational waves are ripples in the universe caused by some of the most energetic cosmic events, from exploding stars to super

  2. #2 peak

    Understanding the causes of magnetic reconnection is important for understanding phenomena around the universe from auroras on Earth, to flares on the surface of the sun, and even to areas surrounding black holes.⏎Next year, MMS will enter Phase 2

  3. #3 peak

    . Each tweet can be a maximum of 280 characters. It is used for a wide variety of things like breaking news, sharing, random conversation and even self-promotion.⏎Pinterest is a social network where you can follow other users.

  4. #4 peak

    numerous food-borne illnesses. These bacteria are especially dangerous for children, pregnant women, and the elderly, and can lead to hospitalization, kidney failure, and even death.⏎Why do people drink raw milk?⏎Some people believe that raw milk tastes

  5. #5 peak

    Although everyone may insist on brushing, not everyone follows the right brushing technique or the frequency of cleaning your pearly whites. Stale food crumbs, fibres and even fruit seeds stuck between teeth and gums result in decay while plaque can further deteriorate

  6. #6 peak

    individual product fragrances, Lavera included. These oils are a much safer alternative to synthetic Fragrance, which have been linked to skin irritation, headaches, and even birth defects.⏎Essential oils are also used as a preservative method in cosmetics

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: "a good time to be a good time to be a good time to be a good time to be a good time to be a good time to be a good time to be a good time to be a good time to be a good

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

    nutritious cereals is wheat bran cereal. It is produced from the outer layer of the wheat grain. The cereal has a rich content of protein, fiber, vitamins

  2. #2 peak

    . Dog bite treatments cost more than a billion dollars each year. The most common<|endoftext|>In 195

  3. #3 peak

    cause a chemical reaction between amino acids and sugars, known as the Maillard reaction. We often encounter it in our everyday life when we fry meat, bake

  4. #4 peak

    sensor on a piece of paper towel coated with water-resistant substance all over, except the areas showing four blood types and a positive

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. L6:531 excites
  2. L3:144 excites
  3. L5:228 suppresses
  4. L5:397 excites
  5. L5:41 excites
  6. L3:604 suppresses
  7. L6:1024 suppresses
  8. L6:757 suppresses
  9. L6:928 excites
  10. L0:984 weakly excites
  11. L5:445 weakly excites
  12. L5:1764 weakly suppresses
  13. L6:380 weakly suppresses
  14. L6:1086 weakly excites
  15. L6:1216 weakly suppresses
  16. L6:400 weakly excites
  17. L2:787 weakly suppresses
  18. L3:52 weakly excites
  19. L4:963 weakly suppresses
  20. L6:786 weakly excites
  21. L3:531 weakly suppresses
  22. L4:2800 weakly suppresses
  23. L2:241 weakly excites
  24. L3:435 weakly excites

Feeds later layers

  1. L9:340 excites
  2. L8:91 suppresses
  3. L9:41 suppresses
  4. L9:128 excites
  5. L9:37 suppresses
  6. L8:77 excites
  7. L11:483 excites
  8. L11:640 suppresses
  9. L9:27 excites
  10. L9:34 excites
  11. L12:1672 excites
  12. L10:349 excites
  13. L10:47 excites
  14. L11:449 excites
  15. L11:1031 excites
  16. L10:3064 excites
  17. L13:2306 excites
  18. L10:1097 suppresses
  19. L8:35 excites
  20. L8:23 excites
  21. L22:130 excites
  22. L8:418 suppresses
  23. L9:620 suppresses
  24. L8:355 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 -2.28 and -2.114 values between -2.11 and -1.958 values between -1.78 and -1.626 values between -1.62 and -1.457 values between -1.45 and -1.2915 values between -1.29 and -1.1334 values between -1.13 and -0.9646 values between -0.96 and -0.8087 values between -0.80 and -0.63139 values between -0.63 and -0.47254 values between -0.47 and -0.30473 values between -0.30 and -0.141040 values between -0.14 and 0.03790 values between 0.03 and 0.19407 values between 0.19 and 0.36247 values between 0.36 and 0.52144 values between 0.52 and 0.6891 values between 0.68 and 0.8545 values between 0.85 and 1.0124 values between 1.01 and 1.1820 values between 1.18 and 1.349 values between 1.34 and 1.5110 values between 1.51 and 1.672 values between 1.84 and 2.001 values between 2.00 and 2.161 values between 2.49 and 2.661 values between 2.82 and 2.991 values between 3.31 and 3.481 values between 4.14 and 4.30offnegativepositiveoften

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. iddle
  2. rawl
  3. ective
  4. agento
  5. ittel
  6. ANGO
  7. Golden
  8. ATEGORIES
  9. ationale
  10. 疾病的

Suppresses

  1. /small
  2. even
  3. -large
  4. alike
  5. .tbl
  6. 甚至是
  7. large
  8. sized
  9. 척
  10. umably