Neuron Explorer

Qwen3-0.6B

'to' in set phrases like 'to come'

A neuron in layer 24 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 word 'to' in fixed phrases such as 'for generations to come', 'here to stay' and 'A Night to Remember'. Turning it up made the writing turn into looping, unclear sentences, so its effect on the output is not easy to read.

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

    lock-style entrances to biometric scanners.⏎Good thing, too. Much of the planet relies on these cloud fortresses, and will do so even more in years to come. Remarkable as the cloud is, there is no reason to think it

  2. #2 peak

    , that will use bacteria that have been genetically engineered to make flu vaccines. But until that day, the egg method and the egg protein in flu vaccines are here to stay.⏎The amount of egg protein varies from one vaccine manufacturer to another. If

  3. #3 peak

    the events that occurred within the region during the Southwestern Campaign.⏎Join t Fight⏎Donate today to preserve Civil War battlefields and the nation’s history for generations to come.<|endoftext|>Understand the True Health Benefits of Almonds⏎What about the

  4. #4 peak

    to be "competitive".⏎Walter Lord, author of the classic "A Night to Remember", describes the process vividly in his 1986 sequel,

  5. #5 peak

    The First Men In The Moon"⏎- 'Star Trek: The Next Generation' - "Rascals"⏎- 'Warehouse 13' - "Time Will Tell"<|endoftext|>I question why "pres" was marked incorrect in this exercise.

  6. #6 peak

    a new international agenda on NCDs.⏎… The aim is for countries to adopt a concise, action-oriented outcome document that will shape the global agendas for generations to come.⏎10 Facts On NCDS⏎1. NCDs account for

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 enough of the set of all set of all set of all set... I can't get enough of it. I can't get enough of it. I can't get enough of it. I can't get enough of

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

    It was also extended toward the future over the following decades, becoming the legal foundation for the reception in Germany of more than 4.5 million so-called “resettlers” (Aussiedler and Spätaussiedler)

  2. #2 peak

    mobilized to respond to a distress call from Vulcan, which was under attack by the Narada. After the destruction of Vulcan, acting captain Spock planned to regroup with the rest of the fleet in the Laurentian system to

  3. #3 peak

    are without a parent or legal guardian). According to the Office of Refugee Resettlement, the average age of these children in 2004 was

  4. #4 peak

    soil. The success of Cuffee’s venture led to the founding, in 1817, of the American Colonization Society, with the mission of relocating freeborn and emancipated blacks to Africa. In 182

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. L23:257 strongly suppresses
  2. L23:505 strongly suppresses
  3. L22:627 strongly suppresses
  4. L20:349 strongly excites
  5. L21:1657 suppresses
  6. L19:3044 suppresses
  7. L21:853 excites 'can' after 'and' or a noun subject
  8. L21:1441 excites
  9. L19:1769 excites
  10. L23:373 suppresses 'to' after a noun or adjective
  11. L19:2491 suppresses
  12. L19:385 suppresses
  13. L20:2802 suppresses
  14. L18:612 excites
  15. L18:1516 suppresses
  16. L20:636 suppresses
  17. L22:1425 excites
  18. L19:2632 suppresses
  19. L21:1663 excites
  20. L18:971 suppresses
  21. L20:1093 suppresses
  22. L17:645 excites
  23. L22:181 suppresses
  24. L18:960 excites

Feeds later layers

  1. L25:503 suppresses
  2. L25:364 suppresses
  3. L27:247 suppresses
  4. L26:296 suppresses
  5. L26:671 suppresses
  6. L26:862 suppresses
  7. L26:675 weakly suppresses
  8. L27:401 weakly suppresses
  9. L26:769 weakly suppresses
  10. L25:181 weakly suppresses
  11. L26:2432 weakly suppresses
  12. L27:2450 weakly suppresses
  13. L26:467 weakly suppresses
  14. L25:600 weakly suppresses
  15. L25:361 weakly suppresses
  16. L26:50 weakly excites
  17. L27:880 weakly suppresses
  18. L27:2609 weakly suppresses
  19. L27:1126 weakly suppresses
  20. L25:672 weakly excites
  21. L25:566 weakly suppresses
  22. L25:108 weakly excites
  23. L26:225 weakly suppresses 'and' joining two abstract nouns
  24. L26:516 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 -14.89 and -14.031 values between -13.17 and -12.301 values between -11.44 and -10.582 values between -9.72 and -8.866 values between -8.86 and -7.995 values between -7.99 and -7.136 values between -7.13 and -6.276 values between -6.27 and -5.4110 values between -5.41 and -4.5515 values between -4.55 and -3.6826 values between -3.68 and -2.8260 values between -2.82 and -1.96100 values between -1.96 and -1.10276 values between -1.10 and -0.242286 values between -0.24 and 0.63892 values between 0.63 and 1.4961 values between 1.49 and 2.3525 values between 2.35 and 3.2123 values between 3.21 and 4.0817 values between 4.08 and 4.9414 values between 4.94 and 5.8018 values between 5.80 and 6.6611 values between 6.66 and 7.528 values between 7.52 and 8.3910 values between 8.39 and 9.257 values between 9.25 and 10.112 values between 10.11 and 10.974 values between 10.97 and 11.843 values between 11.84 and 12.703 values between 12.70 and 13.561 values between 13.56 and 14.422 values between 14.42 and 15.281 values between 15.28 and 16.151 values between 17.01 and 17.873 values between 17.87 and 18.731 values between 18.73 and 19.59offnegativepositiveoften

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. isque
  2. 诱
  3. 筹
  4. 萘
  5. icot
  6. synchron
  7. 阄
  8. sno
  9. endum
  10. unnel

Suppresses

  1. configure
  2. 度过
  3. ujemy
  4. ければ
  5. validate
  6. выбрать
  7. characterize
  8. いる
  9. classify
  10. utilize