Neuron Explorer

Qwen3-0.6B

the word 'value' in the sense of worth

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 responds to the word 'value', meaning worth or importance, as in 'value-added' or 'the value of local products'. Turning it up made the model produce a long run of a single Chinese character, so the steering result tells us little.

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

    based on a single year’s result or small differences among schools and teachers. Texas, for example, rewards schools based on value-added achievement gains calculated using two-stage regression analysis. The state does not hand out

  2. #2 peak

    Cordelia, it is misunderstandings between a father and a daughter that create extreme hatred. The reward of the wicked characters with good in both cases waters down the value of being ethical in life.<|endoftext|>|Model 1: Adapted from G

  3. #3 peak

    competition helps children develop problem solving and other skills, it can be both positive and negative. Negative competition occurs when a child competes for his self-worth and value. This happens especially when parents reinforce the concept that children must "play to win

  4. #4 peak

    2011 > ˅⏎Here are five types of investments states should make in order to implement new teacher evaluation systems successfully.⏎Myths and Facts about Value-Added Analysis November 11, 2011 > ˅

  5. #5 peak

    by developing new connections between local farms and wholesale or institutional buyers,⏎Strong farmer-consumer connections result in a strong local foods system in which consumers identify and appreciate the value of local products, foods are produced and consumed locally, "food miles" and

  6. #6 peak

    comprehensive Spanish help-sheets that have been designed to give you a deeper understanding of the subjects we discuss. You can download a free sample to see if they are value for money. You won’t be disappointed.<|endoftext|>by Rafael Medoff⏎Tur

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: of course, the of course of the of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of of 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

    campaign trail, read the latest polling news, and receive live election results. High school students who love politics are probably already following Mike Allen’s Politico Playbook (Politico) on their iPhones or iPads. Right now the daily news from this

  2. #2 peak

    . He played a vital role in our independence, so we called him as the father of the nation. All our Indians with great love and affection called him as Bapuji. This all info you can write in your Gandhi Jayanti speech

  3. #3 peak

    ama, and Dream Writer. She enjoys meeting young readers and hearing from them what inspires them and how they want to change the world. She works with the Hindu-BusinessLine newspaper, and lives in Chennai.⏎You may also like...⏎- Env

  4. #4 peak

    ). This strike led to them being granted the right to vote. IWD was celebrated for the first time by the United Nations (UN) in 1975. In December 1977, the UN General Assembly adopted a

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. L16:2191 excites
  2. L14:178 suppresses
  3. L6:1915 weakly excites
  4. L3:2634 weakly suppresses
  5. L16:2351 weakly excites
  6. L11:407 weakly excites
  7. L10:299 weakly suppresses
  8. L18:125 weakly suppresses
  9. L18:253 weakly excites
  10. L10:880 weakly suppresses
  11. L8:1616 weakly suppresses
  12. L16:899 weakly excites
  13. L3:1247 weakly suppresses
  14. L6:105 weakly excites
  15. L9:153 weakly excites
  16. L17:765 weakly excites
  17. L1:1857 weakly suppresses
  18. L21:2993 weakly excites
  19. L6:1513 weakly suppresses
  20. L0:1541 weakly excites
  21. L12:2272 weakly suppresses
  22. L1:18 weakly suppresses
  23. L17:754 weakly excites
  24. L2:1834 weakly suppresses

Feeds later layers

  1. L27:651 weakly excites
  2. L25:2383 weakly suppresses
  3. L25:753 weakly suppresses
  4. L25:2605 weakly excites
  5. L26:2518 weakly excites first piece of a split word
  6. L26:711 weakly suppresses 'years' after an age number
  7. L25:2461 weakly excites
  8. L27:2810 weakly suppresses
  9. L25:1422 weakly suppresses
  10. L26:760 weakly suppresses
  11. L27:1038 weakly suppresses
  12. L25:2640 weakly suppresses
  13. L27:435 weakly suppresses
  14. L25:2377 weakly suppresses
  15. L25:1033 weakly suppresses
  16. L25:496 weakly suppresses
  17. L27:2296 weakly suppresses
  18. L27:447 weakly suppresses
  19. L27:2501 weakly suppresses
  20. L26:1954 weakly excites
  21. L27:3065 weakly suppresses
  22. L25:37 weakly suppresses
  23. L25:1802 weakly suppresses
  24. L25:727 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 -10.06 and -8.731 values between -7.41 and -6.081 values between -6.08 and -4.751 values between -4.75 and -3.423 values between -3.42 and -2.0954 values between -2.09 and -0.763689 values between -0.76 and 0.57137 values between 0.57 and 1.909 values between 1.90 and 3.234 values between 3.23 and 4.562 values between 4.56 and 5.882 values between 5.88 and 7.211 values between 7.21 and 8.541 values between 12.53 and 13.861 values between 15.19 and 16.521 values between 41.77 and 43.09offnegativepositiveoften

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. 感
  2. �
  3. 감
  4. 感触
  5. sense
  6. feel
  7. getFile
  8. 感觉
  9. داد
  10. 感人

Suppresses

  1. chner
  2. prites
  3. IDAD
  4. อาย
  5. limited
  6. 殊
  7. olars
  8. idad
  9. ость
  10. iquement