Neuron Explorer

Qwen3-0.6B

comma before a describing clause

A neuron in layer 11 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

The neuron fires on the comma or short word right after the subject of a news-style sentence, just before the extra detail about that subject starts, such as 'The workshops[,] which are funded by...'. Turning it up did not produce any clear topic in the model's output.

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

    show her students live feeds of animals they are learning about so they can see the animals in their native habitats, rather than simply pictures in a textbook.⏎The workshops, which are funded by the Bristol-Myers Squibb Center, are offered

  2. #2 peak

    curl and have turned brown.⏎"I think it has aborted a lot of the acorns. And the leaves turn upside down to keep from losing moisture."⏎The tree, which stands at least 90 feet tall, was estimated by the MU Fore

  3. #3 peak

    31 years at Phoenix Zoo⏎A bald eagle that spent 31 years in the Phoenix Zoo has died, zoo officials announced Wednesday.⏎The bald eagle, named Polaris, was originally rescued in the state of Washington with a broken wing

  4. #4 peak

    -versus-large-truck speed differentials on rural, interstate highways. Sponsored by the U.S. Department of Transportation, commercial trucking companies and private fleets, the study will help regulatory agencies and trucking-company decision-makers establish policies regarding speed limits

  5. #5 peak

    ’s rights activist, author and feminist icon Kamla Bhasin passed away today, September 25, following a brief battle with cancer.⏎Bhasin was a stalwart of the women’s movement in the South Asia region, who spent

  6. #6 peak

    primates following exposure to Delta-9-tetrahydrocannabinol, or THC, the main psychoactive ingredient in marijuana.⏎The nonhuman primates, all of reproductive age with a history of successful conception, received a THC edible once

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: I have been thinking about lately, and I have to do it... something... like... something... to make sure that it's not just a matter of the time... but something... to make sure that it is... as... something...

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

    power amplifier design.⏎The first aim of this text is to fill that need, by providing a detailed guide to the many design decisions that

  2. #2 peak

    coupled with novel molecular dynamics-driven structural refinement to understand the biological roles of higher-order structure in the hepatitis C virus (HCV) RNA genome.⏎Our Specific Aims are designed to reveal numerous new roles for RNA structure in the replication cycle of H

  3. #3 peak

    the mini essay completed in July OR starting afresh, write a detailed report about use of harmony in Haydn’s Joke quartet. Start with the sentence ‘Like the Corelli, the harmony in Haydn’s Joke

  4. #4 peak

    physiography,"⏎says George Zandt, a geophysicist with the University of Arizona in Tucson,⏎and lead author of the study, which appears in today's Nature.⏎"Normally, you have the greatest thickness underneath the tallest mountains

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. L9:2727 excites
  2. L10:451 suppresses
  3. L8:540 weakly suppresses
  4. L8:619 weakly suppresses
  5. L7:1402 weakly suppresses
  6. L10:1858 weakly excites
  7. L10:1389 weakly suppresses
  8. L7:1051 weakly excites
  9. L10:961 weakly suppresses
  10. L8:323 weakly suppresses
  11. L10:920 weakly excites
  12. L7:2222 weakly suppresses
  13. L1:1339 weakly excites
  14. L6:441 weakly suppresses
  15. L0:2356 weakly excites
  16. L7:1165 weakly suppresses
  17. L7:1048 weakly suppresses
  18. L4:350 weakly suppresses
  19. L3:2348 weakly excites
  20. L2:616 weakly excites
  21. L7:2750 weakly suppresses
  22. L7:1678 weakly suppresses
  23. L6:447 weakly suppresses
  24. L7:48 weakly suppresses

Feeds later layers

  1. L12:3032 excites
  2. L13:242 suppresses
  3. L13:2077 excites
  4. L13:1485 suppresses
  5. L14:2244 excites
  6. L13:981 excites
  7. L18:2687 excites
  8. L13:1221 excites comma or verb after a researcher or study subject
  9. L12:1606 excites
  10. L14:353 excites
  11. L15:706 weakly excites
  12. L18:2151 weakly suppresses
  13. L14:1843 weakly excites
  14. L14:1591 weakly excites
  15. L13:2768 weakly excites
  16. L17:2629 weakly excites
  17. L15:332 weakly suppresses
  18. L13:861 weakly excites
  19. L16:2540 weakly excites
  20. L12:2669 weakly excites
  21. L13:2327 weakly excites
  22. L14:607 weakly suppresses
  23. L12:1269 weakly excites
  24. L17:2507 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.

1 values between -3.33 and -3.071 values between -2.03 and -1.781 values between -1.78 and -1.522 values between -1.52 and -1.261 values between -1.26 and -1.0012 values between -1.00 and -0.7426 values between -0.74 and -0.48268 values between -0.48 and -0.222223 values between -0.22 and 0.04939 values between 0.04 and 0.30209 values between 0.30 and 0.5699 values between 0.56 and 0.8242 values between 0.82 and 1.0837 values between 1.08 and 1.3319 values between 1.33 and 1.597 values between 1.59 and 1.854 values between 1.85 and 2.114 values between 2.11 and 2.372 values between 2.37 and 2.631 values between 2.63 and 2.891 values between 2.89 and 3.151 values between 3.15 and 3.411 values between 3.41 and 3.673 values between 3.67 and 3.932 values between 4.70 and 4.961 values between 5.22 and 5.481 values between 6.78 and 7.04offnegativepositiveoften

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. "?⏎⏎
  2. ounced
  3. )!⏎
  4. "?
  5. "..
  6. pine
  7. asty
  8. ”?
  9. !".
  10. asted

Suppresses

  1. orno
  2. ancybox
  3. ucs
  4. InSection
  5. outputPath
  6. uxe
  7. thon
  8. ogs
  9. iciencies
  10. орт