Neuron Explorer

Qwen3-0.6B

unclear: letter fragments inside names

A neuron in layer 4 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 a short letter piece, often a lowercase 'l' or 'i', placed in the middle of a name or word, such as 'Frank-l-inton', 'Gold-il-ocks' or 'immunoglob-ul-ins'. It has no clear meaning beyond that. Steering did not produce a clear change.

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

    of this publication.⏎- Our on-line publications are scanned and captured using Adobe Acrobat. During the capture process some typographical errors may occur. Please contact Sharon Hobrla, email@example.com if you notice any errors which make this publication unusable

  2. #2 peak

    used.⏎Small appliance and electronics transformers may have a thermal cut out built into the winding.⏎Doughnut shaped toroidal transformers are used to save space compared to EI cores, and sometimes

  3. #3 peak

    Columbus, was one of the original 15 settlers of the city. A baker by trade, Heyl came to bake for the soldiers quartered in Franklinton during the War of 1812. He founded the city's

  4. #4 peak

    - 1999⏎African American Nurses have long history with the American Red -⏎Undeterred by initial rejection, Davis persisted and in l9l7 became the⏎first

  5. #5 peak

    ther and Sister A brother drinks from a stream and turns into a deer just when the King and his hunting party ride through the woods.⏎The Frog King.Goldilocks and the Three Bears.More items….⏎Are folktales true stories?⏎

  6. #6 peak

    dietitian, cooking instructor and cookbook author of the newly released "Big Green Cookbook" (Wiley, 2009). Newgent, a Fairlawn, Ohio, native, said she was motivated to write the book after learning

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: the new and the future. The new and the future. The new and the future. The new and the future. The new and the future. The new and the future. The new and the future. The new and the future.

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

    transformations can be specifically assigned for distinct functions such as image classification and segmentation," explains project lead Aydogan Ozcan, professor at the University of California's Samueli School of Engineering, "or they can be dedicated to computing different convolutional filter

  2. #2 peak

    . He practised in several different towns, and became physician, successively, to the Archduke Ferdinand, and to the Emperor Maximilian II.⏎Mattioli's ` Commentarii in sex libros Pedacii Dioscoridis,'

  3. #3 peak

    . The phase of the Moon is calculated daily in the Vedic calendar. Krishna Janmashatmi is celebrated from moonrise to midnight on the tithi of ashtami. Krishna Janmashtami is also known as Krishna Jay

  4. #4 peak

    optical pyrometer has a range of measuring temperature of 1400oC to near about 3500oC.<|endoftext|>The “presidi” translates as “garrisons” (from the French word, “to

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. L1:31 suppresses
  2. L2:549 suppresses
  3. L2:1097 suppresses
  4. L0:1224 weakly suppresses
  5. L3:1376 weakly suppresses
  6. L0:425 weakly excites
  7. L0:1301 weakly suppresses
  8. L0:178 weakly suppresses
  9. L3:2461 weakly suppresses
  10. L2:448 weakly suppresses
  11. L1:491 weakly excites
  12. L0:2873 weakly excites
  13. L3:59 weakly excites
  14. L2:75 weakly suppresses
  15. L1:45 weakly excites
  16. L1:2683 weakly excites
  17. L1:2362 weakly suppresses
  18. L2:1036 weakly excites
  19. L0:113 weakly suppresses
  20. L1:179 weakly suppresses
  21. L2:1536 weakly suppresses
  22. L0:116 weakly suppresses
  23. L1:994 weakly suppresses
  24. L2:2102 weakly suppresses

Feeds later layers

  1. L18:499 excites
  2. L25:1965 excites the digit 1 in age ranges
  3. L17:512 excites
  4. L5:142 excites
  5. L5:90 excites
  6. L6:1151 weakly excites
  7. L15:632 weakly suppresses
  8. L16:2259 weakly suppresses
  9. L18:2499 weakly suppresses
  10. L6:2953 weakly excites
  11. L5:340 weakly suppresses zeros inside long numbers
  12. L6:293 weakly excites
  13. L6:1831 weakly excites
  14. L8:880 weakly excites
  15. L18:785 weakly suppresses
  16. L6:236 weakly excites
  17. L25:2688 weakly suppresses
  18. L23:1661 weakly suppresses
  19. L20:416 weakly excites the "1" in years 191x
  20. L25:2170 weakly suppresses
  21. L7:9 weakly excites
  22. L7:2271 weakly suppresses
  23. L18:706 weakly excites
  24. L10:1025 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 -0.48 and -0.461 values between -0.44 and -0.423 values between -0.39 and -0.374 values between -0.37 and -0.353 values between -0.35 and -0.338 values between -0.33 and -0.3019 values between -0.30 and -0.2815 values between -0.28 and -0.2641 values between -0.26 and -0.2444 values between -0.24 and -0.2263 values between -0.22 and -0.19132 values between -0.19 and -0.17136 values between -0.17 and -0.15187 values between -0.15 and -0.13263 values between -0.13 and -0.10297 values between -0.10 and -0.08357 values between -0.08 and -0.06382 values between -0.06 and -0.04369 values between -0.04 and -0.02417 values between -0.02 and 0.01303 values between 0.01 and 0.03222 values between 0.03 and 0.05178 values between 0.05 and 0.07145 values between 0.07 and 0.1093 values between 0.10 and 0.1283 values between 0.12 and 0.1452 values between 0.14 and 0.1630 values between 0.16 and 0.1919 values between 0.19 and 0.2110 values between 0.21 and 0.2311 values between 0.23 and 0.255 values between 0.25 and 0.286 values between 0.28 and 0.303 values between 0.30 and 0.323 values between 0.34 and 0.361 values between 0.36 and 0.392 values between 0.39 and 0.41offnegativepositiveoften

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. environmentally
  2. ",{
  3. spb
  4. ��
  5. �示
  6. ��
  7. 躬
  8. ude
  9. "),"
  10. �

Suppresses

  1. dm
  2. pun
  3. 名额
  4. FileType
  5. book
  6. ticking
  7. licher
  8. اما
  9. nim
  10. phot