Neuron Explorer

Qwen3-0.6B

first digit after 'more than'

A neuron in layer 3 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 lights up on the first digit of a number that follows phrases like 'more than' or 'the last' or 'past', such as the 1 in '10 percent' or '1,100 passengers'. When it was turned up, the model's output filled with numbers and counting, though the text was not coherent.

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

    had managed to sustain such high long-term growth and transformation. Frankly, I was stumped. I have spent over four decades in Asia, the last 16 years living in China, and no one had ever asked me that question.⏎

  2. #2 peak

    04, when the region experienced a period of heavy cloud cover. Those clouds, which resemble convective clouds on Earth, could have rained more than 10 centimet

  3. #3 peak

    in 2017 emissions coincided with the IEA’s release of its long-range energy forecast and its first in-depth China analysis in the past 10 years.⏎In its 2017 Global Carbon Budget, the scientists

  4. #4 peak

    the spread of the disease and the value of Georgia’s crop was reduced more than 10 percent. Nearly every variety grown back then was susceptible to the virus and a

  5. #5 peak

    a rumor mill that a black pregnant woman had killed by officers. Columbia Avenue in North Philadelphia erupted into a three-day riot, leaving hundreds injured and more than 200 restaurants and retail stores destroyed. The retail corridor, later renamed Cecil B

  6. #6 peak

    continued to Auckland, New Zealand, where two of the passengers – a couple in their 70's – were airlifted to hospital. More than 1,100 passengers have been asked to undergo blood tests.⏎Hepatitis

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: I can't get to the point of a person who is the same as I am. So the time. So I can think. So the time. So I can think. So the time. So I can think. So the time

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

    ⏎|Day 1 – Critical Reading Single Passage||3 hours||CR Assessment homework|⏎|Day 2 – Dual & Short Passages, & Sentence Comp.||3 hours||Single Passage Assessment|⏎|Day 3 – The Essay||

  2. #2 peak

    pull up stakes⏎6: wander from a direct or straight course [syn: sidetrack⏎English → English⏎, v. i. [imp. & p. p. Departed⏎; p. pr. & vb.⏎.]

  3. #3 peak

    – these thinkers are focused on issues too deep for other kids to understand, & may be prone to depression as they ponder the meaning of life. Careers involving philosophy & spirituality

  4. #4 peak

    , courgettes⏎3. WHEN? (see crop chart)⏎Right time: temperature, how long til harvest? Seed packets will tell you when to sow &

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. L2:986 suppresses
  2. L2:1769 suppresses
  3. L2:1665 suppresses
  4. L2:2015 excites
  5. L2:2089 suppresses
  6. L2:36 weakly excites
  7. L1:39 weakly excites
  8. L1:1816 weakly excites
  9. L2:1494 weakly suppresses
  10. L2:1715 weakly excites
  11. L2:1334 weakly suppresses
  12. L2:672 weakly suppresses line breaks before page-section headings
  13. L1:1375 weakly suppresses
  14. L2:1955 weakly excites
  15. L2:1438 weakly suppresses
  16. L2:111 weakly suppresses
  17. L1:1728 weakly excites
  18. L0:722 weakly suppresses
  19. L2:1437 weakly suppresses
  20. L0:2620 weakly excites
  21. L2:925 weakly excites
  22. L2:483 weakly suppresses
  23. L0:1610 weakly excites
  24. L2:1356 weakly excites

Feeds later layers

  1. L6:1478 suppresses
  2. L10:214 suppresses
  3. L9:2517 suppresses
  4. L5:2840 suppresses period after abbreviation in citations
  5. L7:1853 suppresses
  6. L4:789 weakly excites
  7. L25:652 weakly suppresses
  8. L7:114 weakly suppresses
  9. L24:1098 weakly excites
  10. L5:1813 weakly suppresses
  11. L5:2422 weakly excites
  12. L6:676 weakly suppresses
  13. L12:1340 weakly suppresses
  14. L7:883 weakly suppresses
  15. L27:1564 weakly suppresses
  16. L25:486 weakly suppresses
  17. L21:514 weakly suppresses
  18. L7:765 weakly suppresses
  19. L10:86 weakly suppresses
  20. L8:1214 weakly excites
  21. L25:764 weakly suppresses
  22. L10:639 weakly suppresses
  23. L12:1334 weakly suppresses
  24. L5:978 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 -1.99 and -1.871 values between -1.76 and -1.641 values between -1.30 and -1.184 values between -0.61 and -0.4914 values between -0.49 and -0.3878 values between -0.38 and -0.26376 values between -0.26 and -0.151199 values between -0.15 and -0.031645 values between -0.03 and 0.08465 values between 0.08 and 0.2076 values between 0.20 and 0.3114 values between 0.31 and 0.436 values between 0.43 and 0.548 values between 0.54 and 0.663 values between 0.66 and 0.773 values between 0.77 and 0.892 values between 0.89 and 1.002 values between 1.12 and 1.231 values between 1.23 and 1.353 values between 1.46 and 1.583 values between 1.69 and 1.811 values between 1.92 and 2.041 values between 2.04 and 2.151 values between 2.49 and 2.61offnegativepositiveoften

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. zsche
  2. nette
  3. .toByteArray
  4. toUpperCase
  5. Stamford
  6. osphere
  7. ляет
  8. publié
  9. slaughtered
  10. Warfare

Suppresses

  1. nicer
  2. favour
  3. liest
  4. natur
  5. Jvm
  6. ording
  7. dàng
  8. yrs
  9. just
  10. 俱