Neuron Explorer

Qwen3-0.6B

'1' in 21st or 11th, before 'st'

A neuron in layer 25 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 digit '1' in ordinal numbers like '21st' or '11th', just before the 'st' ending. It is negative on the 'i' in 'fiftieth' and 'seventieth'. Turned up, the model produced long runs of 'st', which matches what it tends to predict next.

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

    . The first modern vacuum cleaner. The first aeroplane. The vacuum tube diode. Bakelite.⏎And what about the first decade of the 21st century? The list is long, and you may be surprised by some of

  2. #2 peak

    put enormous strain on the planet’. The United States stood 17th in the rankings.⏎Meanwhile, among neighboring South Asian countries, India ranked 131st on ‘planetary pressures-adjusted’ HDI, Bangladesh, 1

  3. #3 peak

    , Pelican ©2012, 978-1-4556-1637-4⏎In the 21st century, with GPS and vast improvements in diving equipment, many wrecks are

  4. #4 peak

    nearby town of New Alexandria to learn the blacksmith trade before enlisting in the Union Army.⏎In 1861, William enlisted in the Pennsylvania 11th Reserve Infantry. He was just 18 and small in stature.

  5. #5 peak

    indigenous to the area. Most states have<|endoftext|>A new smartphone app aims to provide a cheaper alternative to ultrasound in Africa by bringing an old technique into the 21st century.⏎"We couldn't hear anything,”

  6. #6 peak

    human activities can theoretically exert some influence.(7) As the 21st century began, some experts thought it plausible that the Sun might have driven part

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: how do you like the future? How do you like the future? How do you like the future? How do you like the future? How do you like the future? How do you like the future? How do you like the future

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

    about the implications for the Geneva Refugee Convention on its seventieth anniversary.⏎The postwar refugee situation affected many people who had lost their homes

  2. #2 peak

    of designing for resilience and is critical for climate change mitigation and adaptation.<|endoftext|>On July 20, 2019 we will celebrate the fiftieth anniversary of the Apollo 11 moon landing. People of my generation thought

  3. #3 peak

    restaurant area dedicated to the exploits of these two intrepid airmen, was so named because it opened in 1969, on the fiftieth anniversary of the flight. Highly recommended for a visit by people, such as

  4. #4 peak

    moon.⏎- Ashura: 10th Muharram*⏎Ashura is an Arabic word meaning “ten”. It is celebrated on the 10th day of the Hejrah-e qamari month Muharram

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. L19:2587 excites
  2. L22:1055 excites
  3. L21:327 suppresses
  4. L24:118 excites
  5. L19:118 suppresses
  6. L19:395 excites
  7. L21:2433 excites
  8. L21:870 excites
  9. L7:2157 suppresses 'the' after 'like' or 'stay'
  10. L3:1316 weakly suppresses
  11. L16:118 weakly suppresses
  12. L21:1741 weakly suppresses
  13. L17:166 weakly suppresses
  14. L23:779 weakly excites
  15. L17:1818 weakly excites
  16. L20:1687 weakly excites
  17. L21:708 weakly suppresses
  18. L24:364 weakly excites 'at' in 'at least' and 'most' or 'least' qualifiers
  19. L24:1762 weakly suppresses
  20. L19:723 weakly excites
  21. L14:392 weakly suppresses
  22. L20:78 weakly excites
  23. L10:50 weakly excites
  24. L22:332 weakly suppresses

Feeds later layers

  1. L27:792 strongly excites
  2. L27:2850 excites
  3. L26:3019 excites
  4. L27:824 excites
  5. L27:903 excites
  6. L27:183 excites
  7. L27:359 excites
  8. L26:729 weakly excites
  9. L26:728 weakly excites
  10. L26:261 weakly excites
  11. L26:2836 weakly excites
  12. L27:268 weakly excites
  13. L26:1016 weakly excites
  14. L26:2996 weakly excites
  15. L27:1488 weakly excites
  16. L27:285 weakly excites
  17. L26:685 weakly excites
  18. L26:3056 weakly suppresses
  19. L26:2256 weakly suppresses
  20. L26:1008 weakly suppresses
  21. L27:632 weakly excites
  22. L26:2761 weakly suppresses
  23. L27:409 weakly excites
  24. L26:1234 weakly 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.

1 values between -18.47 and -16.882 values between -12.09 and -10.501 values between -10.50 and -8.914 values between -4.13 and -2.53178 values between -2.53 and -0.943602 values between -0.94 and 0.6636 values between 0.66 and 2.2520 values between 2.25 and 3.849 values between 3.84 and 5.4415 values between 5.44 and 7.036 values between 7.03 and 8.637 values between 8.63 and 10.222 values between 10.22 and 11.813 values between 11.81 and 13.414 values between 13.41 and 15.004 values between 15.00 and 16.592 values between 16.59 and 18.191 values between 18.19 and 19.781 values between 21.38 and 22.972 values between 22.97 and 24.562 values between 24.56 and 26.161 values between 26.16 and 27.751 values between 27.75 and 29.341 values between 29.34 and 30.941 values between 35.72 and 37.311 values between 38.91 and 40.501 values between 43.69 and 45.28offnegativepositiveoften

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. st
  2. rd
  3. nd
  4. st
  5. sti
  6. sts
  7. sth
  8. (st
  9. ST
  10. stab

Suppresses

  1. ikes
  2. icularly
  3. bitten
  4. 独角
  5. Unidos
  6. .unsplash
  7. inin
  8. orian
  9. -dismissible
  10. 久了