Neuron Explorer

Qwen3-0.6B

end of document boundaries

A neuron in layer 12 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 invisible marker the model uses to show that one web page or document has ended and another begins. It is a weak signal compared with the strongest neurons. Steering it did not change the text in a recognisable way.

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

    intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of qualified health providers with questions you may have regarding medical conditions.<|endoftext|>The Amendment Process and the African American Fight for Representation and Equality⏎Black History Month

  2. #2 peak

    one thing which was enjoyable and fascinating on your youngster, however you're fearful about whether or not the fabric will really enhance your kid's long-term college efficiency.<|endoftext|>“When Paroh will say to you, “Provide for yourself a mofait

  3. #3 peak

    of the capital.⏎The contents of this article is licensed from www.wikipedia.org under the GNU Free Documentation License. Click here to see the transparent copy and copyright details<|endoftext|>What is Celiac / Coeliac Disease?Tweet⏎What is Celiac Disease

  4. #4 peak

    |⏎|Launched:||25 October 1941|⏎|Commissioned:||28 April 1942|⏎|Decommissioned:||28 June 1945|⏎|Renamed:

  5. #5 peak

    positives can be reduced by focusing on high-risk patients and using regular instead of digital mammogram's.⏎All content © Copyright 2000 - 2013 WorldNow and WXOW. All Rights Reserved.⏎Persons with disabilities who

  6. #6 peak

    also hope to take measurements<|endoftext|>RWJF Priority: Use pricing strategies—both incentives and disincentives—to

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: what if the otherworldly beings who are more advanced than the ones who have been studying the stars, and who have been fascinated with the concept of time, and the concept of the stars? What is the way to understand the balance 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

    preserving their industrial past. This was one of our favorite areas to explore.⏎9 thoughts on “The Lachine Canal & Canada Malting Silos”⏎The place looks good

  2. #2 peak

    become overweight and obese, we must attack the problem from all angles. Maybe the day will come when gyms are as easily found as fast food restaurants."⏎More information: Beat Schuler,

  3. #3 peak

    human experience. Her works invite readers to contemplate the beauty of the natural world and their place in it, encouraging a deeper understanding of oneself and the surrounding environment.⏎Q2: How does Mary<|endoftext|>Unified Modeling Language⏎<language> (UML

  4. #4 peak

    Front Ecol Environ, http://www.joeroman.com/wordpress/wp-content/uploads/2014/09/roman-frontiers.pdf.<|endoftext|>Oil & Natural Gas Projects⏎Transmission, Distribution, & Refining⏎Mult

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. L9:375 suppresses
  2. L7:332 suppresses
  3. L8:1090 excites
  4. L10:283 suppresses
  5. L9:690 excites
  6. L11:379 excites
  7. L10:1038 excites
  8. L7:503 excites
  9. L11:1377 excites
  10. L10:2821 excites
  11. L8:362 excites unclear: evaluative or explanatory phrase words
  12. L10:2475 excites
  13. L9:1459 excites
  14. L8:848 weakly excites
  15. L11:2715 weakly suppresses end of a document
  16. L7:542 weakly suppresses
  17. L10:928 weakly suppresses
  18. L7:301 weakly suppresses
  19. L8:385 weakly excites
  20. L8:1193 weakly excites
  21. L11:440 weakly suppresses
  22. L11:629 weakly excites
  23. L10:1298 weakly excites
  24. L7:567 weakly suppresses

Feeds later layers

  1. L14:2827 suppresses
  2. L13:969 excites
  3. L14:2045 suppresses
  4. L14:571 excites
  5. L14:3011 suppresses
  6. L13:34 excites
  7. L15:2990 excites
  8. L14:1069 suppresses
  9. L13:1358 suppresses
  10. L13:2182 excites
  11. L13:1201 excites
  12. L18:2322 suppresses end-of-text token after a finished passage
  13. L14:2803 suppresses
  14. L14:2113 excites
  15. L16:2476 excites
  16. L17:1980 suppresses
  17. L14:1574 suppresses
  18. L14:889 suppresses
  19. L15:457 excites
  20. L13:381 excites
  21. L14:1628 weakly suppresses
  22. L13:1589 weakly suppresses
  23. L14:2367 weakly excites
  24. L17:397 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.

3 values between -1.30 and -1.232 values between -1.23 and -1.172 values between -1.17 and -1.103 values between -1.10 and -1.042 values between -1.04 and -0.973 values between -0.97 and -0.916 values between -0.84 and -0.783 values between -0.78 and -0.7110 values between -0.71 and -0.6513 values between -0.65 and -0.5811 values between -0.58 and -0.5221 values between -0.52 and -0.4522 values between -0.45 and -0.3846 values between -0.38 and -0.3262 values between -0.32 and -0.25133 values between -0.25 and -0.19211 values between -0.19 and -0.12479 values between -0.12 and -0.061378 values between -0.06 and 0.01890 values between 0.01 and 0.07356 values between 0.07 and 0.14128 values between 0.14 and 0.2057 values between 0.20 and 0.2727 values between 0.27 and 0.3421 values between 0.34 and 0.407 values between 0.40 and 0.475 values between 0.47 and 0.532 values between 0.60 and 0.661 values between 0.66 and 0.731 values between 0.86 and 0.922 values between 0.99 and 1.051 values between 1.25 and 1.32offnegativepositiveoften

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. stad
  2. -now
  3. elta
  4. (Local
  5. 惧
  6. Lotto
  7. แหละ
  8. combined
  9. ora
  10. laat

Suppresses

  1. rint
  2. ior
  3. .SetString
  4. ure
  5. aki
  6. )⏎⏎⏎
  7. uggestions
  8. idia
  9. :⏎⏎⏎
  10. íd