Neuron Explorer

Qwen3-0.6B

unclear: 'no matter' and 'several' before reasons

A neuron in layer 23 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

The neuron fires on a few different words, mostly "no" in "no matter what" and quantity words like "several" or "few" that come before a list of ways or reasons. It is hard to find one clear theme. Turning it down made the text repeat the word "who" and turning it up gave wandering text.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    15, 5, 2.5, 2, and 1.8° per step.⏎Step angle = 360°/No of steps per revolution⏎Where m = Number of Phase winding on the

  2. #2 peak

    the info and together with the real-time checksum authentication which the ZFS file system performs on all drives, you will never have to be concerned about losing any information no matter what.<|endoftext|>also a strong stimulator of creativity.⏎It became clear that

  3. #3 peak

    product, check<|endoftext|>Help support New Advent and get the full contents of this website as an instant download or CD-ROM. Includes the Catholic Encyclopedia, Church Fathers, Summa, Bible and more all for only $19.99...⏎A

  4. #4 peak

    Feelings of frustration⏎* Feelings of incompetence⏎* Feelings of loneliness⏎* Feeling of powerlessness⏎* Believing that things will turn out bad no matter what action is taken⏎* Feelings of low self-esteem⏎* Bel

  5. #5 peak

    when the value rises. Shorting has a high potential for loss, so it’s important to understand the risks.⏎Cryptocurrencies can be shorted in a few different ways, and it’s good for investors to learn how to short them,

  6. #6 peak

    too far afield; but it is clear from all that is known of the<|endoftext|>Obese & Fat German Shepherd⏎Like humans, dogs can become obese for several reasons, and the effect is just as severe in them. German Shepherds

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: that to have a who-in-the to and to and to and to and to and to and to and to and to and and in and to and and in and to and and in and to and and in and and to and and in

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

    Eukratides I of Bactria, a land in central Asia (Afghanistan) that was conquered by Alexander. The adoption of the title 'Great King', implying imperial rather than merely royal status, followed some years after Euk

  2. #2 peak

    several meters wide – floating in the Great Pacific Garbage Patch. These ghost nets are known to be a major hazard to

  3. #3 peak

    945.”⏎Associate Professor of the Department of Philosophy, Zhanna Simonova, told the students about the contribution of the Kursk Medical Institute to the Great Victory: “In 1941, the first graduation of doctors of

  4. #4 peak

    compounds have helped to protect the personal health and safety of humans, they create risks and concerns to the environment that are constantly being researched.”⏎An investigative study by the Associated Press in 2008 discovered that a vast array of pharmaceuticals were

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. L18:1265 suppresses
  2. L18:602 suppresses
  3. L4:731 excites
  4. L17:2348 excites
  5. L3:914 excites "any" in negated sentences
  6. L11:233 suppresses
  7. L4:326 suppresses
  8. L10:167 weakly suppresses
  9. L22:2882 weakly suppresses
  10. L13:2114 weakly suppresses
  11. L16:1433 weakly excites
  12. L20:1917 weakly suppresses
  13. L20:2278 weakly suppresses
  14. L16:1333 weakly excites 'two' after 'divided into' or 'there are'
  15. L0:973 weakly excites
  16. L5:1146 weakly excites
  17. L18:635 weakly excites
  18. L21:20 weakly suppresses
  19. L15:2272 weakly suppresses
  20. L16:1851 weakly suppresses
  21. L9:1363 weakly excites ordinal numbers like 16th or 1st
  22. L6:972 weakly excites "well" after "how"
  23. L7:232 weakly excites
  24. L22:404 weakly excites

Feeds later layers

  1. L24:258 weakly excites
  2. L24:35 weakly excites
  3. L26:86 weakly excites
  4. L24:2538 weakly suppresses
  5. L24:158 weakly excites
  6. L25:544 weakly suppresses
  7. L24:2753 weakly suppresses
  8. L24:1263 weakly suppresses
  9. L27:1151 weakly suppresses
  10. L24:2914 weakly suppresses
  11. L25:112 weakly suppresses
  12. L25:1404 weakly suppresses
  13. L27:440 weakly excites
  14. L24:1080 weakly suppresses
  15. L27:1941 weakly excites
  16. L24:116 weakly excites
  17. L27:146 weakly excites
  18. L24:148 weakly suppresses
  19. L25:1224 weakly excites
  20. L26:144 weakly suppresses
  21. L24:111 weakly excites
  22. L25:3045 weakly suppresses
  23. L25:1135 weakly excites
  24. L27:1651 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.

2 values between -14.16 and -13.411 values between -10.38 and -9.622 values between -9.62 and -8.861 values between -8.86 and -8.102 values between -8.10 and -7.352 values between -7.35 and -6.593 values between -6.59 and -5.834 values between -5.83 and -5.078 values between -5.07 and -4.3210 values between -4.32 and -3.5615 values between -3.56 and -2.8027 values between -2.80 and -2.0434 values between -2.04 and -1.29424 values between -1.29 and -0.532743 values between -0.53 and 0.23360 values between 0.23 and 0.9992 values between 0.99 and 1.7548 values between 1.75 and 2.5035 values between 2.50 and 3.2619 values between 3.26 and 4.0213 values between 4.02 and 4.7817 values between 4.78 and 5.538 values between 5.53 and 6.299 values between 6.29 and 7.053 values between 7.05 and 7.813 values between 7.81 and 8.562 values between 8.56 and 9.328 values between 9.32 and 10.084 values between 10.08 and 10.843 values between 10.84 and 11.602 values between 12.35 and 13.111 values between 13.87 and 14.632 values between 14.63 and 15.381 values between 15.38 and 16.14offnegativepositiveoften

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. оч
  2. recent
  3. 原因之一
  4. AccessType
  5. 部分
  6. 小
  7. oto
  8. 东西
  9. itim
  10. 主要

Suppresses

  1. 会觉得
  2. 在内的
  3. етесь
  4. 所说
  5. 何况
  6. 所做的
  7. who
  8. ersion
  9. 您好
  10. saja