Neuron Explorer

Qwen3-0.6B

glosses, translations and unit conversions

A neuron in layer 16 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 fires on the small opening mark just before a translation, a meaning or a unit conversion, such as the quote before "hemp" or the bracket before "22,841 ft". It goes strongly negative on the digit 9 in years like 1933 or 1930. Turning it up did not produce a clear, related theme.

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

    to be used for canvas sails, ropes and rigging of these vessels. The word ‘canvas’ originated from the Arabic word for cannabis, a Latin word meaning ‘hemp‘. All of the British captains carried hemp seed in their stores to

  2. #2 peak

    would manage other plants.⏎Looking after non-bonsai plants involves re-potting them frequently to allow the root systems to expand, but bonsai cultivation is the opposite.⏎Since you will need to keep your bonsai tree small, you have

  3. #3 peak

    ” is by strengthening whatever muscles (sh<|endoftext|>The baguette cut is named due to its long, rectangular shape after the French word baguette, for “long rod”, coming down from the Latin baculum, meaning “a stick”.

  4. #4 peak

    book credibly debunks the move to tax or cap carbon dioxide emissions both from technical and political perspectives.⏎The word “neurotoxic” literally translates as “nerve poison”. Broadly described, a

  5. #5 peak

    gyne is different with its marked compact, low and bushy stature (5-10 cm) and small green leaves (each stem is 3.4 cm wide). During planting in the aquarium, the longest upright shoots should

  6. #6 peak

    are the highest mountains of each of the seven continents.⏎Aconcagua, South America⏎With its cone-shaped peak rising at 6,962 m (22,841 ft), Cerro Aconcagua holds the title

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: her name was a little different. She was a little... her... her... her... her... her... her... her... her... her... her... her... her... her... her... her... her... her... her...

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

    6 EO-1, Salyut 7 EP-5. More...⏎Makarov Makarov, Oleg Grigoryevich (1933-20

  2. #2 peak

    atta Facts⏎Mohammad Hatta (1902-1980), one of the foremost intellectuals in the Afro

  3. #3 peak

    final manned flight of the Skylab space station in 1973 and 1974.⏎He was born January 23, 1930, in Okemah, Oklahoma. He received a Bachelor of Science

  4. #4 peak

    the process that people go through when formulating solutions to problems.⏎Herbert Simon was born in Milwaukee, Wisconsin, on June 15th, 1916. He went to the Milwaukee public school system,

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. L9:345 excites
  2. L14:1302 suppresses
  3. L12:2180 suppresses
  4. L10:2612 excites birth dates, the day before the year
  5. L13:2500 excites
  6. L12:2830 suppresses
  7. L11:1895 excites
  8. L14:2058 suppresses
  9. L15:711 suppresses
  10. L14:2267 excites
  11. L14:412 excites
  12. L12:1461 excites
  13. L12:2720 excites
  14. L11:3043 suppresses
  15. L15:2869 suppresses
  16. L12:1279 excites
  17. L15:1942 excites
  18. L14:867 excites
  19. L15:400 suppresses
  20. L11:3049 excites
  21. L10:728 suppresses
  22. L15:2069 suppresses
  23. L11:1663 excites
  24. L12:55 excites

Feeds later layers

  1. L17:2105 excites
  2. L18:2063 excites
  3. L17:164 suppresses
  4. L19:2176 suppresses
  5. L17:1888 suppresses
  6. L19:2895 suppresses
  7. L20:2256 excites
  8. L17:2749 suppresses
  9. L17:2356 excites
  10. L17:2051 excites
  11. L17:3058 excites
  12. L19:2715 excites
  13. L17:1704 excites
  14. L17:1625 excites
  15. L17:2921 weakly suppresses
  16. L19:2541 weakly excites
  17. L17:2733 weakly excites
  18. L20:2867 weakly suppresses
  19. L20:2389 weakly suppresses
  20. L17:2189 weakly excites
  21. L17:2922 weakly suppresses
  22. L21:1580 weakly suppresses
  23. L18:3003 weakly excites
  24. L18:2235 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 -5.32 and -5.151 values between -4.66 and -4.491 values between -3.83 and -3.661 values between -3.50 and -3.331 values between -3.16 and -3.002 values between -3.00 and -2.831 values between -2.83 and -2.671 values between -2.67 and -2.501 values between -2.33 and -2.175 values between -2.17 and -2.002 values between -2.00 and -1.844 values between -1.84 and -1.675 values between -1.67 and -1.5110 values between -1.51 and -1.3414 values between -1.34 and -1.1717 values between -1.17 and -1.0129 values between -1.01 and -0.8454 values between -0.84 and -0.6866 values between -0.68 and -0.51147 values between -0.51 and -0.34320 values between -0.34 and -0.181012 values between -0.18 and -0.011602 values between -0.01 and 0.15419 values between 0.15 and 0.32119 values between 0.32 and 0.4948 values between 0.49 and 0.6512 values between 0.65 and 0.824 values between 0.82 and 0.987 values between 0.98 and 1.152 values between 1.15 and 1.32offnegativepositiveoften

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. MotionEvent
  2. ussen
  3. aneous
  4. utely
  5. 月中
  6. ysql
  7. aneously
  8. onitor
  9. OLID
  10. ức

Suppresses

  1. 性格
  2. 个性
  3. 的性格
  4. pedigree
  5. unique
  6. karakter
  7. characteristics
  8. Profile
  9. 特质
  10. egal