Neuron Explorer

Qwen3-0.6B

numbers and list openers, mixed

A neuron in layer 20 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 is loosely tied to numbers: it fires on digits and number tokens in arithmetic, years and totals. It goes quiet or negative on colons that introduce a list. Turning it up or down gave text that did not clearly follow this pattern.

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

    equivalent? Yes, they are.⏎Now we can add the cardinal numbers and the numerators. First the cardinal numbers. That is 2 + 3 = 5. And then the numerators. 1 + 1 = 2

  2. #2 peak

    in school to increase from 75 to 88 percent. MENA achieved the same increase in 2 decades: from 1990 to 2010.2⏎However, it is important to note that regional

  3. #3 peak

    3. Pope Paul III, who convoked the Council, oversaw the first eight sessions (1545–47), while the twelfth to sixteenth sessions (1551–52) were overseen by

  4. #4 peak

    3 constructed in 1938 have been abandoned. They were located in Iverson Park. The present system consists of Wells No. 4 through 10, two elevated storage tanks, one ground level reservoir and a distribution system

  5. #5 peak

    early 1900s. The construction of the hotel began in 1901 and was completed in just two years, in 1903. The hotel was designed by Harrison Albright, an architect known for his

  6. #6 peak

    umes, animal manure, compost and crop residues. Organic nutrients are slowly released as materials decompose. Compost and manure provide additional benefits, such as increasing organic matter which contributes to building healthy soil.⏎Why don’t all farmers use nutrients

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 first and most commonly used in the ingraph, the most commonly used in the second 2022 research head in the first reheptetand (1) (2) The second head in

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

    to calculate and return the volume of a sphere based on the given radius.⏎- Your second subroutine should be called getVolumeOfCuboid(), take three parameters called width, length and height and apply the relevant formula to calculate and return the volume

  2. #2 peak

    Vitamin K Deficiency⏎- Food Sources of Vitamin K⏎In the early 20th century, it was believed that humans only required four types of nutrients: proteins, carbohydrates, fats, and minerals. Since then, our understanding of health

  3. #3 peak

    also packed with vitamins and minerals, fiber and an extensive amount of important phytochemicals.⏎What are carbohydrates?⏎Carbohydrates are molecules consisting of three elements: carbon, hydrogen and oxygen. They fuel the body with the energy it needs for

  4. #4 peak

    form what we call our solar system. The Earth is a planet. I'm sure you knew that... There are in fact 9 major planets around the sun: Mercury, Venus, the Earth, Mars, Jupiter, Saturn, Uranus,

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. L19:2463 suppresses
  2. L10:2827 weakly suppresses
  3. L9:1860 weakly excites
  4. L10:1437 weakly suppresses
  5. L7:525 weakly suppresses
  6. L19:2794 weakly excites
  7. L14:1311 weakly suppresses
  8. L8:496 weakly suppresses
  9. L16:804 weakly suppresses
  10. L4:2871 weakly excites
  11. L12:1858 weakly excites
  12. L18:486 weakly excites
  13. L18:2892 weakly excites
  14. L18:874 weakly suppresses
  15. L7:1065 weakly excites
  16. L9:1955 weakly excites
  17. L17:1880 weakly excites
  18. L2:1942 weakly excites
  19. L18:2135 weakly excites
  20. L10:1723 weakly suppresses
  21. L17:264 weakly suppresses
  22. L12:674 weakly excites
  23. L18:1434 weakly suppresses
  24. L9:1326 weakly excites

Feeds later layers

  1. L22:1545 weakly suppresses
  2. L21:2331 weakly suppresses
  3. L22:685 weakly excites
  4. L21:634 weakly excites
  5. L21:2175 weakly excites
  6. L22:1099 weakly suppresses
  7. L21:2397 weakly excites
  8. L25:460 weakly suppresses
  9. L21:2660 weakly suppresses
  10. L25:1282 weakly excites
  11. L25:2785 weakly suppresses
  12. L22:559 weakly suppresses
  13. L21:1523 weakly excites
  14. L21:1485 weakly suppresses
  15. L22:1826 weakly suppresses
  16. L24:976 weakly suppresses
  17. L22:485 weakly excites
  18. L24:2538 weakly excites
  19. L23:1457 weakly excites
  20. L22:9 weakly excites
  21. L21:1056 weakly suppresses
  22. L26:1878 weakly suppresses
  23. L21:84 weakly excites the word 'according'
  24. L21:915 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 -9.27 and -8.851 values between -7.16 and -6.731 values between -6.73 and -6.311 values between -6.31 and -5.891 values between -5.46 and -5.042 values between -5.04 and -4.621 values between -4.62 and -4.191 values between -4.19 and -3.771 values between -2.92 and -2.502 values between -2.50 and -2.073 values between -2.07 and -1.659 values between -1.65 and -1.2316 values between -1.23 and -0.80225 values between -0.80 and -0.381832 values between -0.38 and 0.041554 values between 0.04 and 0.47224 values between 0.47 and 0.8922 values between 0.89 and 1.322 values between 1.32 and 1.744 values between 1.74 and 2.161 values between 2.59 and 3.011 values between 3.01 and 3.431 values between 3.86 and 4.281 values between 4.70 and 5.131 values between 7.24 and 7.67offnegativepositiveoften

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. acular
  2. usty
  3. irate
  4. atee
  5. azzi
  6. 功劳
  7. 酱
  8. @Table
  9. inal
  10. 而来

Suppresses

  1. صف
  2. ложение
  3. lists
  4. pint
  5. Demand
  6. спис
  7. tere
  8. -response
  9. .response
  10. ЛЕ