Neuron Explorer

Qwen3-0.6B

the letter 'l' starting a word piece

A neuron in layer 26 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 letter 'l' when it begins the second half of a word that has been split into pieces, as in 'lily', 'lizards' and 'lice'. The steering samples did not show a clear theme, so the link to this habit is only a guess.

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

    entions for Racial and Ethnic Minorities<|endoftext|>Lily of the valley, or convallaria, is not a true lily like the Easter or Asiatic lily, but is still toxic to dogs, humans and other animals. If your

  2. #2 peak

    —making it a natural match for interior growing conditions. It also likes to be potbound and will bloom even in a relatively tiny pot.⏎Common name: Fire lily, bush lily, clivia⏎Botanical name: Clivia mini

  3. #3 peak

    iles, frog, etc⏎- Dead animals without slaughtering based on Islamic law⏎- Repulsive animals like spiders, rats, mice, maggots, lice, etc⏎- An animal that are not slaughtered according to Islamic law⏎

  4. #4 peak

    even expressed the slightest appetite for food. They are companions whose kin are also welcome: their lowbush cousins, their huckleberry relatives<|endoftext|>For these Australian lizards, baby’s first steps are a sprint!⏎This

  5. #5 peak

    a few of the events that you should definitely attend? There’s the Wild World of Animals that features 14 interesting animals—from scorpions to legless lizards—sponsored by N.A.T.U.R.E., Pace’s Environmental Club

  6. #6 peak

    .⏎Remove dried stems only after foliage dies back.⏎Rainy Side Notes⏎Lilium 'White Stargazer' is the most popular of the white upright lilies and used by many for forcing. In fact, Peter Schenk bred it

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: afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword afterword

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

    by women from being performed in churches. Despite this setback, she persevered in the ways she could, including hosting musical soirées that included guests such as Wolfgang Amadeus Mozart.⏎Her music, though largely considered out of fashion by the end

  2. #2 peak

    expanding to address international issues and global concerns.⏎Warren County's Legacy for Mexico's Border Maquiladoras, 1 Golden Gate U. Envtl. L.J.<|endoftext|>In a world where most teenagers are focused on Tweets, Facebook and

  3. #3 peak

    topmost layer of earth. Aeration should be considered for all high-traffic areas, lawns with excess thatch, and clay soil.⏎Reasons to Aerate your Lawn⏎Compact soil features severely limited pore space that would otherwise contain air

  4. #4 peak

    global average temperatures rise, researchers are beginning to see some indications that the areas where certain pathogens can live are shifting, he said. [5 Ways Climate Change Will Affect Your Health]⏎"We are seeing the spread of pathogens to new ecological niches

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. L23:2833 excites
  2. L18:102 suppresses
  3. L18:778 excites
  4. L23:1760 suppresses
  5. L16:880 suppresses
  6. L16:95 excites
  7. L24:2164 excites
  8. L17:188 excites 'a' before a noun phrase
  9. L20:174 excites
  10. L20:901 suppresses
  11. L21:252 weakly excites
  12. L21:1072 weakly suppresses
  13. L16:117 weakly suppresses
  14. L19:299 weakly suppresses
  15. L22:1383 weakly excites
  16. L18:687 weakly suppresses
  17. L10:33 weakly excites
  18. L19:2088 weakly suppresses
  19. L18:420 weakly excites
  20. L7:520 weakly suppresses
  21. L6:328 weakly suppresses
  22. L11:2859 weakly suppresses
  23. L19:183 weakly suppresses
  24. L5:1294 weakly excites

Feeds later layers

  1. L27:564 weakly suppresses
  2. L27:1712 weakly suppresses
  3. L27:856 weakly excites
  4. L27:1532 weakly suppresses
  5. L27:2276 weakly suppresses
  6. L27:1248 weakly suppresses
  7. L27:689 weakly excites
  8. L27:10 weakly suppresses
  9. L27:2231 weakly suppresses
  10. L27:2699 weakly suppresses
  11. L27:282 weakly excites
  12. L27:227 weakly excites
  13. L27:1862 weakly suppresses
  14. L27:1182 weakly excites
  15. L27:1152 weakly excites
  16. L27:2092 weakly suppresses
  17. L27:1750 weakly excites
  18. L27:1540 weakly suppresses
  19. L27:1147 weakly excites
  20. L27:2577 weakly excites
  21. L27:317 weakly excites
  22. L27:465 weakly suppresses
  23. L27:2420 weakly suppresses
  24. L27:1718 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 -23.67 and -22.481 values between -20.08 and -18.892 values between -18.89 and -17.692 values between -16.50 and -15.301 values between -15.30 and -14.103 values between -11.71 and -10.524 values between -10.52 and -9.323 values between -9.32 and -8.121 values between -8.12 and -6.9310 values between -6.93 and -5.7318 values between -5.73 and -4.5315 values between -4.53 and -3.3427 values between -3.34 and -2.14125 values between -2.14 and -0.952815 values between -0.95 and 0.25670 values between 0.25 and 1.4581 values between 1.45 and 2.6441 values between 2.64 and 3.8419 values between 3.84 and 5.0316 values between 5.03 and 6.2310 values between 6.23 and 7.437 values between 7.43 and 8.6211 values between 8.62 and 9.822 values between 9.82 and 11.026 values between 11.02 and 12.215 values between 12.21 and 13.411 values between 14.60 and 15.804 values between 15.80 and 17.002 values between 17.00 and 18.192 values between 18.19 and 19.392 values between 22.98 and 24.17offnegativepositiveoften

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. olate
  2. ole
  3. ouples
  4. oles
  5. apa
  6. ega
  7. im
  8. ounty
  9. Engl
  10. rie

Suppresses

  1. 猗
  2. пуска
  3. ypos
  4. 指
  5. ippet
  6. пад
  7. pró
  8. .pointer
  9. vis
  10. tbl