Neuron Explorer

Qwen3-0.6B

the word 'according'

A neuron in layer 21 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 strongly on the word 'according', as in 'according to the laws' or 'according to legend'. Turning it up or down did not bring out this word, and the output was mostly unrelated maths and weather text.

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

    , He which soweth sparingly shall reap also sparingly; and he which soweth bountifully shall reap also bountifully.⏎Every man according as he purposeth in his heart, so let him give; not grudging

  2. #2 peak

    have offended, and only asked the King not to make him languish long.⏎Thereupon some said he was a traitor, others that he should be judged according to the laws he had made, which were so sanguinary that often words

  3. #3 peak

    the acceptance or rejection of this Constitution, an election shall be held on the second Monday of November, in the year 1857, to be conducted according to existing laws regulating the election of Delegates in Congress, so far as applicable

  4. #4 peak

    events he refused to attend⏎the General Assembly on 14 November 1883 arguing the action against him as⏎“unconstitutional and illegal” according to the laws of the Church and⏎defiantly set sail for Scotland

  5. #5 peak

    Since men are neither omniscient nor infallible, they must be free to agree or disagree, to cooperate or to pursue their own independent course, each according to his own rational judgment. Freedom is the fundamental requirement of man’s mind.⏎

  6. #6 peak

    start to cry”⏎; “The train departs at noon”⏎, go away⏎] [ant: come⏎2: be at variance with; be out of line with [syn: deviate⏎] [ant: conform⏎3:

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 number of the company and the number of the company's business. Now, for the company's business, it has been organized in the start of the first step, and then the next step is to take a further step in the next

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

    Duke of Mantua, who was afterwards Viceroy of Navarre and Valencia, and who recommended him to Philip II in the last year of that monarch's reign. Philip appointed him grand historiographer (cronista mayor) of America

  2. #2 peak

    A New, Authentic and Complete Collection of Voyages Round the World…Complete Historical Account of Captain Cook's First, Second, Third and Last Voyages… Edited by George William Anderson and published

  3. #3 peak

    People with Disabilities⏎As children with developmental disabilities get older, they will face many changes in each phase of their life, with the most significant arguably being the transition to adulthood. Transitions may look like moving towns, leaving the family

  4. #4 peak

    the newspaper shows a photo of Trump and Kim Jong Un on its front page.⏎Trump and Kim were asked by reporters about the possibility of opening a liaison office during their meeting last month in Hanoi, Vietnam. “It’s actually not a bad

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. L18:1889 excites
  2. L8:2686 weakly suppresses
  3. L20:1498 weakly excites
  4. L16:2125 weakly excites
  5. L8:168 weakly suppresses
  6. L20:1115 weakly excites
  7. L10:3006 weakly excites
  8. L19:2443 weakly excites
  9. L10:432 weakly suppresses
  10. L20:362 weakly excites
  11. L18:1190 weakly excites
  12. L20:1934 weakly excites
  13. L15:2378 weakly suppresses
  14. L18:1623 weakly suppresses
  15. L16:1543 weakly excites
  16. L7:899 weakly suppresses
  17. L15:65 weakly excites
  18. L18:1017 weakly suppresses
  19. L12:704 weakly suppresses
  20. L14:919 weakly suppresses
  21. L1:87 weakly excites
  22. L9:863 weakly suppresses
  23. L0:206 weakly suppresses
  24. L10:1573 weakly excites

Feeds later layers

  1. L22:2125 weakly excites
  2. L22:3041 weakly suppresses
  3. L22:382 weakly excites
  4. L22:1800 weakly suppresses
  5. L22:932 weakly excites
  6. L22:2906 weakly suppresses
  7. L22:1311 weakly excites
  8. L22:1460 weakly suppresses
  9. L25:3065 weakly suppresses
  10. L23:1054 weakly excites
  11. L24:1841 weakly excites
  12. L24:601 weakly excites
  13. L23:2664 weakly suppresses
  14. L25:2952 weakly excites
  15. L22:1544 weakly excites
  16. L23:645 weakly excites
  17. L23:722 weakly excites
  18. L22:2344 weakly excites number after 'account for' or 'over'
  19. L22:1806 weakly excites "to" introducing a purpose clause
  20. L24:2403 weakly excites
  21. L22:670 weakly excites
  22. L24:1511 weakly suppresses
  23. L22:2413 weakly suppresses
  24. L26:1944 weakly excites

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 -15.13 and -14.452 values between -14.45 and -13.771 values between -11.73 and -11.061 values between -11.06 and -10.385 values between -10.38 and -9.708 values between -9.70 and -9.029 values between -9.02 and -8.347 values between -8.34 and -7.6613 values between -7.66 and -6.9916 values between -6.99 and -6.3135 values between -6.31 and -5.6335 values between -5.63 and -4.9562 values between -4.95 and -4.2789 values between -4.27 and -3.59138 values between -3.59 and -2.92178 values between -2.92 and -2.24296 values between -2.24 and -1.56361 values between -1.56 and -0.88536 values between -0.88 and -0.20918 values between -0.20 and 0.48494 values between 0.48 and 1.16313 values between 1.16 and 1.83147 values between 1.83 and 2.51121 values between 2.51 and 3.1961 values between 3.19 and 3.8730 values between 3.87 and 4.5514 values between 4.55 and 5.239 values between 5.23 and 5.902 values between 5.90 and 6.582 values between 6.58 and 7.262 values between 7.26 and 7.941 values between 9.30 and 9.971 values between 11.33 and 12.01offnegativepositiveoften

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. Charts
  2. tables
  3. público
  4. 吃什么
  5. 门户网站
  6. cig
  7. Tate
  8. ilan
  9. chemicals
  10. 橱柜

Suppresses

  1. st
  2. ocratic
  3. .bel
  4. ู้
  5. _ST
  6. �行
  7. ystone
  8. _step
  9. 导演
  10. utzer