Neuron Explorer

Qwen3-0.6B

the digit 1 starting a number

A neuron in layer 26 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 when it is the first character of a longer number, such as the 1 in 10,000 or 1,887 meters. When it was turned up, the model fell into repeating the word 'step', which fits the fact that it nudges the model toward words like 'step' and 'layer'.

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

    1-24.⏎USDA zones: 5-9.⏎Heat zones: 9-5.⏎Height: 4-10 inches (10-25 cm).⏎Two to three-inch diameter spherical umbel of

  2. #2 peak

    's heat shield after it was shed about two and a half minutes before the rover landed. In the photo, the heat shield is roughly 50 feet (16 meters) away from the spacecraft, and had just been separated about three seconds

  3. #3 peak

    However, some people don't look like they're dead at all. Read on!⏎Tutankhamun lived from 1332 BC to 1323 BC. He is widely regarded as King Tut. When his tomb

  4. #4 peak

    26 GMT, 1:26 p.m. PDT), after falling free from its parent aircraft while flying 39,000 feet (11,887 meters) above the Pacific Ocean.⏎"It was nominal,

  5. #5 peak

    of other bodily tissues for fuel, such as fat and muscle. Every athlete is different, but most agree that it’s best to eat a carbohydrate-rich meal 1 to 3 hours before the big race to give the food time to digest and

  6. #6 peak

    a long and rich history, dating back to prehistoric times. Early examples of Japanese art include Jomon pottery, which was created during the Jomon period (10,000 BCE to 300 BCE), and Hani

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: what if we can machines discover colors? colors that are not only colors but colors that can produce colors! Imagine colors that can influence the colors of the spaces! Imagine colors that can create colors! Imagine colors that can influence colors! Imagine colors

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

    bits, or qubits, can encode both at the same time. This essentially enables quantum computers to work through multiple scenarios simultaneously, rather than exploring them one after the other. However, these mixed states don’t last long, so

  2. #2 peak

    help and Operations Management homework help. Live tutors are available for 24x7 hours helping students in their Characteristics of Linear Programming related problems. We provide step by step Characteristics of Linear Programming question's answers with 100% plagiarism free

  3. #3 peak

    Triangles Exercise 6.5 are given here. There are several questions in this exercise. Our subject expert teachers have solved each of these questions in a step by step manner. All these solutions are beneficial for the students of class 10

  4. #4 peak

    money, they have to think outside of the box. Since the music became a digital product record sales give the wall to the royalties. The audience is changing step by step, now teenagers present a new type of people, they are “digital natives

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. L25:1601 strongly excites
  2. L25:374 strongly suppresses
  3. L21:1888 suppresses
  4. L25:437 excites
  5. L23:465 excites
  6. L25:390 suppresses
  7. L23:731 excites
  8. L23:79 excites
  9. L22:1233 excites
  10. L19:2469 suppresses
  11. L23:1373 excites
  12. L22:260 suppresses
  13. L20:889 excites
  14. L22:811 suppresses
  15. L24:268 excites
  16. L22:653 suppresses
  17. L22:2089 excites
  18. L21:1730 suppresses
  19. L21:518 excites
  20. L20:602 suppresses
  21. L18:2629 suppresses
  22. L25:407 excites
  23. L21:971 excites
  24. L25:284 suppresses

Feeds later layers

  1. L27:687 excites
  2. L27:421 excites
  3. L27:726 suppresses
  4. L27:588 excites
  5. L27:551 excites
  6. L27:815 excites
  7. L27:209 suppresses
  8. L27:666 excites
  9. L27:1939 excites
  10. L27:521 weakly excites
  11. L27:1106 weakly excites
  12. L27:533 weakly excites
  13. L27:729 weakly suppresses
  14. L27:706 weakly excites
  15. L27:1504 weakly suppresses
  16. L27:122 weakly excites
  17. L27:434 weakly excites
  18. L27:2379 weakly excites
  19. L27:931 weakly excites
  20. L27:1615 weakly suppresses
  21. L27:96 weakly suppresses
  22. L27:335 weakly excites
  23. L27:2658 weakly excites
  24. L27:2945 weakly suppresses capital P in acronyms

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 -21.72 and -20.601 values between -12.74 and -11.621 values between -8.25 and -7.138 values between -7.13 and -6.017 values between -6.01 and -4.896 values between -4.89 and -3.7611 values between -3.76 and -2.6456 values between -2.64 and -1.521444 values between -1.52 and -0.401437 values between -0.40 and 0.73436 values between 0.73 and 1.85185 values between 1.85 and 2.97128 values between 2.97 and 4.0957 values between 4.09 and 5.2259 values between 5.22 and 6.3421 values between 6.34 and 7.4619 values between 7.46 and 8.587 values between 8.58 and 9.719 values between 9.71 and 10.835 values between 10.83 and 11.953 values between 11.95 and 13.072 values between 13.07 and 14.191 values between 15.32 and 16.441 values between 16.44 and 17.562 values between 17.56 and 18.681 values between 22.05 and 23.17offnegativepositiveoften

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. step
  2. layer
  3. step
  4. row
  5. chapter
  6. term
  7. line
  8. (step
  9. day
  10. element

Suppresses

  1. станов
  2. enames
  3. اث
  4. cabinets
  5. 累累
  6. createForm
  7. 澪
  8. جموع
  9. defaultMessage
  10. umes