Neuron Explorer

Qwen3-0.6B

quantity words like 'one', 'more', 'amount of'

A neuron in layer 19 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on words that measure how much or how long, such as 'one', 'more', 'amount of' or 'percent of', right before the thing being measured. It goes slightly negative on words like 'at' or 'a' in plain physical descriptions. Turning it up made the model repeat phrases about the newspaper or temperature, which does not clearly match the pattern.

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

    converts rotary motion into linear motion.⏎The ‘lead’ of a Lead Screw is the distance which a Lead Screw Nut will move during one 360° rotation of the screw. A smaller Lead results in a

  2. #2 peak

    repeating event per unit time. It is also referred to as temporal frequency, which emphasizes the contrast to spatial frequency and angular frequency. The period is the duration of one cycle in a repeating event, so the period is the reciprocal of the frequency.⏎

  3. #3 peak

    through which these beliefs have been given to us, but also a guide for our present worries and undertakings, is not so common. It takes a lot more reflection (and life experience) to realize the rules and ideas about behavior we have

  4. #4 peak

    in thermal runaway and over-current conditions. Protection circuitry is straightforward and easy to implement but adds to the bill-of-materials (BoM) and takes up more space

  5. #5 peak

    body of harmful toxins.<|endoftext|>Object-oriented editing means the ability to edit and/or process individual objects, such as a MIDI or audio clip — whether they last the entire length of a song or only a few milliseconds. This is a different way of

  6. #6 peak

    proportional fonts, in which different characters have different widths.⏎In the monospaced example above, a wide character (like “m”) takes up the same amount of space as a thin character (like “i”).⏎Monospaced type looks

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: when I have a day's worth of work, and the day's worth of work is not enough, you have to put up a day's worth of work in a month's worth of days worth of months, or you have to be

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

    RF) waves are a form of electromagnetic energy. They move through space at the speed of light and can be man-made or occur naturally. They move through space at the speed of light and can be man-made or occur naturally

  2. #2 peak

    molecular clock.⏎DNA is the giant molecule in each of our cells that holds the architectural plan of our body and physical functions. Over the generations, DNA changes at a presumably steady rate. For birds, that's about 2% change (for

  3. #3 peak

    unit types are: DECIMAL, SCIENTIFIC, ENGINEERING, ARCHITECTURAL, FRACTIONAL. The ‘precision’ section allows you to pick the number of decimal places for your dimensions. If you are doing a project for

  4. #4 peak

    Countess of Rosebery, died, at Dalmeny House, in Scotland, at the age of 39. As the only child of Meyer de Ams

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. L17:1693 suppresses
  2. L17:2607 suppresses
  3. L18:1412 excites
  4. L15:97 suppresses
  5. L15:1521 excites
  6. L14:258 excites
  7. L18:1348 suppresses
  8. L18:1176 weakly suppresses
  9. L18:1315 weakly excites
  10. L17:673 weakly excites
  11. L18:1304 weakly suppresses
  12. L18:573 weakly suppresses
  13. L17:2588 weakly excites
  14. L17:1895 weakly suppresses
  15. L16:576 weakly excites
  16. L18:948 weakly excites
  17. L16:2968 weakly suppresses
  18. L16:556 weakly excites
  19. L17:209 weakly excites
  20. L18:618 weakly excites
  21. L13:1978 weakly excites
  22. L12:602 weakly suppresses
  23. L18:2718 weakly excites
  24. L18:2846 weakly suppresses

Feeds later layers

  1. L22:1310 excites
  2. L20:988 excites
  3. L20:2734 weakly excites
  4. L20:695 weakly excites
  5. L23:161 weakly suppresses
  6. L20:731 weakly excites
  7. L25:2969 weakly excites
  8. L25:1286 weakly suppresses
  9. L27:335 weakly excites
  10. L20:1038 weakly suppresses
  11. L20:599 weakly suppresses
  12. L20:581 weakly suppresses
  13. L23:1247 weakly suppresses
  14. L27:2452 weakly excites
  15. L21:908 weakly suppresses
  16. L22:660 weakly excites
  17. L21:2851 weakly suppresses
  18. L27:1780 weakly suppresses
  19. L22:1679 weakly excites
  20. L21:708 weakly suppresses
  21. L21:2171 weakly excites
  22. L24:1368 weakly suppresses
  23. L22:2563 weakly excites formal document and standards citations
  24. L20:2482 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 -6.41 and -6.091 values between -5.78 and -5.462 values between -5.46 and -5.151 values between -5.15 and -4.831 values between -4.83 and -4.524 values between -4.52 and -4.215 values between -4.21 and -3.897 values between -3.89 and -3.588 values between -3.58 and -3.2610 values between -3.26 and -2.9517 values between -2.95 and -2.6332 values between -2.63 and -2.3229 values between -2.32 and -2.0156 values between -2.01 and -1.6963 values between -1.69 and -1.38118 values between -1.38 and -1.06187 values between -1.06 and -0.75272 values between -0.75 and -0.43560 values between -0.43 and -0.121021 values between -0.12 and 0.20529 values between 0.20 and 0.51313 values between 0.51 and 0.82217 values between 0.82 and 1.14146 values between 1.14 and 1.4587 values between 1.45 and 1.7760 values between 1.77 and 2.0844 values between 2.08 and 2.4038 values between 2.40 and 2.7123 values between 2.71 and 3.0219 values between 3.02 and 3.347 values between 3.34 and 3.659 values between 3.65 and 3.977 values between 3.97 and 4.285 values between 4.28 and 4.602 values between 4.60 and 4.913 values between 4.91 and 5.231 values between 5.54 and 5.853 values between 5.85 and 6.17offnegativepositiveoften

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. tape
  2. liga
  3. Pager
  4. �
  5. crib
  6. igar
  7. bubb
  8. transformer
  9. transformers
  10. Spreadsheet

Suppresses

  1. 年年底
  2. otland
  3. 年底
  4. .distance
  5. oted
  6. _bonus
  7. engeance
  8. 郢
  9. _FAMILY
  10. offseason