Neuron Explorer

Qwen3-0.6B

the 'n' that starts nectar

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the letter 'n' when it begins the word 'nectar', and it was also strong on a few other word starts. Turning it up made the model repeat 'stay' and similar words, which has little to do with flowers, so the link between the two is unclear.

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

    'the little hanger', and is thought to come from the way the sunbirds and sugarbirds appear to 'hang' from the flowers while feeding on the nectar.⏎Three formerly separate species have been included in this complex, namely

  2. #2 peak

    .⏎Crocus are a good choice to attract bees in the early spring. They're also pollinated by butterflies.⏎Asters are perennials that provide nectar and pollen, and do well when planted in late summer and fall.⏎Ger

  3. #3 peak

    Dictionary and Cyclopedia⏎- n. A sea-duck of the subfamily Fuligulinœ and genus Clangula; a garrot. The common goldeneye is C. glaucion

  4. #4 peak

    resource might have been premature. He thinks that boll weevils’ consumption of pollen from host plants other than cotton might actually occur as they’re feeding on nectar. He says that the nutritional properties of such pollens and their effect on

  5. #5 peak

    ily.⏎Lorikeets have a very simple digestive system. Proteins and fats are obtained from pollen and insects, and carbohydrates come from the sugars found in nectar. They spend a large amount of time foraging for food, which indicates

  6. #6 peak

    O-deyt ](verb): to make bare; strip; denude.⏎This fun word comes from the latin word “denudatus” which means “naked” or “ bare.”⏎With the days getting colder, the deciduous

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: swapping swap swapping swap swapping swapping swapping swap swapping swap swapping swap swapping swapping swapping swapping swapping swapping swapping swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping swap swapping

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

    and life of the people of the time. It was written in the second century A.D. by Illango Adikal, the prince of the royal lineage who renounced the royal heritage and took to the robes.⏎In Silapathikaram

  2. #2 peak

    estone mansions and residences forming an urban district that is a conservation area and one of the most outstanding Renaissance urban districts in France.⏎Bar-le-Duc invites you to explore its magnificent upper town<|endoftext|>Transdisciplinary Environmental Science

  3. #3 peak

    Pakistan, affecting both students and staff. From unwanted physical contact to verbal abuse, sexual harassment can take many forms and can have severe psychological and emotional consequences for those who experience it. Unfortunately, despite efforts to raise awareness and implement preventative measures, instances

  4. #4 peak

    . Prove AQ = AD.⏎At the time of writing the hour and minute hands of my clock are at⏎right angles. How long will it be before they

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. L16:55 excites
  2. L22:2299 excites
  3. L17:391 weakly excites
  4. L21:567 weakly suppresses the word 'in' after a verb or noun
  5. L20:931 weakly excites the digit 6 inside years
  6. L23:491 weakly excites
  7. L18:127 weakly excites
  8. L18:924 weakly suppresses
  9. L19:617 weakly excites
  10. L19:277 weakly suppresses
  11. L16:673 weakly suppresses
  12. L9:2554 weakly suppresses
  13. L20:2883 weakly suppresses
  14. L19:1286 weakly excites
  15. L0:1145 weakly excites
  16. L19:770 weakly excites
  17. L16:74 weakly suppresses
  18. L11:981 weakly suppresses
  19. L8:369 weakly suppresses 'of' after 'learned' or 'because'
  20. L22:3065 weakly excites
  21. L2:2028 weakly excites
  22. L24:319 weakly suppresses
  23. L0:771 weakly suppresses
  24. L25:2670 weakly excites

Feeds later layers

  1. L27:946 weakly excites
  2. L27:3056 weakly suppresses
  3. L27:2140 weakly excites
  4. L27:435 weakly suppresses
  5. L27:1199 weakly suppresses
  6. L27:2151 weakly excites
  7. L27:1853 weakly suppresses
  8. L27:1783 weakly excites
  9. L27:369 weakly suppresses
  10. L27:1717 weakly excites 'st' starting a word fragment
  11. L27:1012 weakly suppresses
  12. L27:1828 weakly excites
  13. L27:1182 weakly suppresses
  14. L27:1200 weakly suppresses
  15. L27:992 weakly excites
  16. L27:2276 weakly excites
  17. L27:1655 weakly suppresses
  18. L27:2594 weakly excites
  19. L27:3028 weakly excites
  20. L27:2716 weakly suppresses
  21. L27:2354 weakly excites
  22. L27:2678 weakly suppresses
  23. L27:2447 weakly suppresses
  24. L27:2114 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 -26.33 and -24.811 values between -18.75 and -17.242 values between -17.24 and -15.722 values between -15.72 and -14.211 values between -14.21 and -12.693 values between -12.69 and -11.185 values between -11.18 and -9.6610 values between -9.66 and -8.155 values between -8.15 and -6.6316 values between -6.63 and -5.1237 values between -5.12 and -3.6053 values between -3.60 and -2.08272 values between -2.08 and -0.573192 values between -0.57 and 0.95159 values between 0.95 and 2.4657 values between 2.46 and 3.9832 values between 3.98 and 5.4918 values between 5.49 and 7.0116 values between 7.01 and 8.528 values between 8.52 and 10.043 values between 10.04 and 11.554 values between 11.55 and 13.071 values between 13.07 and 14.582 values between 14.58 and 16.103 values between 16.10 and 17.612 values between 17.61 and 19.131 values between 19.13 and 20.641 values between 22.16 and 23.681 values between 32.77 and 34.28offnegativepositiveoften

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. stay
  2. stays
  3. staying
  4. remain
  5. work
  6. stay
  7. remains
  8. oth
  9. stayed
  10. remain

Suppresses

  1. .swap
  2. swap
  3. swapping
  4. ierce
  5. swap
  6. ature
  7. ASSERT
  8. 自有
  9. Zack
  10. liche