Neuron Explorer

Qwen3-0.6B

number right after 'about' or 'less than'

A neuron in layer 18 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 gap just before a quantity, as in 'about [ ]160 square miles' or 'less than [ ]0.01%'. It is waiting for a number to come next. Turning it up made the model produce long runs of digits.

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

    , 6–8⏎A hunk of ice about the size of the state of Connecticut has started to collapse in Antarctica. According to satellite images, about 160 square miles of the Wilkins Ice Shelf shows massive cracking, scientists

  2. #2 peak

    of the total global population). Population ageing is happening more quickly than in the past. For example, while France had almost 150 years to adapt to a change from 10% to

  3. #3 peak

    solution to the problem of the labour costs incurred for moving the bottles; however, they require a high initial investment, while the remuage process will last about 7 days. A winery using these methods might not be able to meet a

  4. #4 peak

    The ocean’s breadth and depth remain the most extensive, unexplored habit.⏎Activity 2 – Freshwater Habitats⏎Freshwater Habitats cover less than 0.01 % of the earth’s

  5. #5 peak

    and younger adults. (Courneya, Estabrooks & Nigg 1997) Unlike attitude, which tends to consistently account for approximately 30% of behavioral intention, social norms are rarely even significant and always account

  6. #6 peak

    ’s spiral arms are.”⏎Named GLIMPSE360 (Galactic Legacy Mid-Plane Survey Extraordinaire project), the deep infrared survey captures only about 3% of the sky, but because it focuses on the plane of the Milky

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: the idea that, rather than "to" rather than "to" instead of "to" the "to" at the end in the "to" "to" "to" "to" "to" "to" "

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

    . Despite being high in fat and sodium, tamales are a rich source of vitamins, minerals and fiber that support optimal health.⏎A 100-gram serving of tamales contains about 18.

  2. #2 peak

    the whaling industry, which nearly drove some species to extinction. At its peak in 1856, the United States whaling industry produced between four- and five-million gallons of whale oil annually.⏎Whale-oil lamps used one

  3. #3 peak

    2, WGBH Media Library & Archives, accessed March 29, 2017, http://openvault.wgbh.org/catalog/V_4E4EE3EAD24043E2BC0

  4. #4 peak

    ’s in order to win a prize, that the problem was solved.⏎Joan Marie Galat, Pelican ©2012, 978-1-4556-1637-4⏎In the

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. L13:923 suppresses word right before a number or year
  2. L4:53 excites blank slot before a number
  3. L17:285 suppresses
  4. L10:220 suppresses hyphen inside compound words
  5. L6:344 excites
  6. L5:2539 excites
  7. L16:256 excites number right after 'age of' or 'ages'
  8. L8:111 suppresses the space before a year in dates
  9. L0:602 excites words right before a number
  10. L4:1129 excites
  11. L4:490 weakly suppresses decade digit in years like 1930s
  12. L6:17 weakly suppresses
  13. L14:348 weakly suppresses
  14. L16:723 weakly suppresses
  15. L4:2691 weakly excites
  16. L4:2129 weakly suppresses
  17. L2:422 weakly excites
  18. L5:2991 weakly suppresses
  19. L14:140 weakly suppresses
  20. L1:349 weakly suppresses
  21. L17:382 weakly suppresses closing bracket after a fact in parentheses
  22. L17:525 weakly suppresses
  23. L16:7 weakly suppresses
  24. L4:2532 weakly excites surname pieces split across tokens

Feeds later layers

  1. L27:170 excites unclear: spaces in mixed-language text
  2. L21:339 excites spot just before a year
  3. L25:33 excites spot where a number is due
  4. L22:824 excites blank before an expected number
  5. L22:318 suppresses
  6. L21:302 excites
  7. L26:103 excites blank space before a number
  8. L27:283 excites the spot right before a number
  9. L20:131 excites
  10. L19:26 suppresses
  11. L21:1358 suppresses
  12. L23:77 excites the space right before a number
  13. L25:253 suppresses
  14. L26:611 suppresses
  15. L19:43 suppresses
  16. L27:306 suppresses
  17. L21:189 excites document break token mid-sentence
  18. L20:61 suppresses
  19. L27:2172 excites
  20. L22:372 excites empty spot before a number or "3D"
  21. L19:58 suppresses
  22. L20:4 excites
  23. L25:182 excites
  24. L20:72 suppresses

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.

331 values between -2.23 and -1.223020 values between -1.22 and -0.22513 values between -0.22 and 0.797 values between 0.79 and 1.804 values between 1.80 and 2.811 values between 2.81 and 3.821 values between 4.83 and 5.831 values between 5.83 and 6.841 values between 6.84 and 7.851 values between 10.87 and 11.881 values between 11.88 and 12.892 values between 16.92 and 17.931 values between 18.94 and 19.951 values between 19.95 and 20.961 values between 20.96 and 21.961 values between 22.97 and 23.981 values between 24.99 and 26.002 values between 26.00 and 27.001 values between 27.00 and 28.012 values between 28.01 and 29.025 values between 29.02 and 30.031 values between 32.05 and 33.053 values between 33.05 and 34.064 values between 34.06 and 35.072 values between 37.09 and 38.09offnegativepositiveoften

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. 理会
  2. filib
  3. 里的
  4. 的世界
  5. Chili
  6. ︰
  7. roscope
  8. 月中
  9. Cunning
  10. Boca

Suppresses

  1. ger
  2. acious
  3. -
  4. push
  5. ao
  6. 报警
  7. burst
  8. ins