Neuron Explorer

Qwen3-0.6B

end of text token, weak

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 weakly when a document seems to finish, right at the hidden end-of-text marker. Turning it up made the model write long, looping lines of unrelated text, so the effect is hard to read.

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

    style developed in full. Her main source of inspiration was Mexican folk culture, which she blended with elements from pre-Colombian and Catholic mythology in her paintings.<|endoftext|>|Wednesday, 31 December 2003

  2. #2 peak

    carried out for the samples with greenschist discontinuities, and found the characteristic of shear creep failure is closely related to the development of the structure surface.<|endoftext|>Japan’s Aerospace Exploration Agency (JAXA) is charting new territory in

  3. #3 peak

    :<|endoftext|>Copyright © University of Cambridge. All rights reserved.⏎'Aim High' printed from

  4. #4 peak

    :<|endoftext|>The hamstring muscles – the biceps femoris, semitendinosus,

  5. #5 peak

    , it checks to see if the current wiki page (containing the web bug) has changed and needs re-indexing.⏎On the first release containing the indexer<|endoftext|>Publications and Tools⏎Comprehensive School Reform and Student Achievement: A Meta-

  6. #6 peak

    2-year-old has three gardens, which he uses primarily to grow tomatoes, sweet banana peppers and okra. The work varies in intensity during the growing season.<|endoftext|>Iocasta: What is an exile's life? Is it great misery?⏎

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 problem of language and the process of learning in the classroom" is something that can be studied in more depth at the universities in the world, as a model for the learners' education in the world. As the students in the

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

    full article here: http://ourgame.mlblogs.com/2014/08/18/out-at-home/⏎- Related link: Click here to view

  2. #2 peak

    um or spreads from somewhere else (colon⏎Ovarian Epithelial, Fallopian Tube, and Primary Peritoneal Cancer …⏎(7 days ago) Treatment of ovarian epithelial, fallopian tube, and primary peritone

  3. #3 peak

    Students explore the Orange Revolution. In this Orange Revolution lesson, students examine the climate of the Ukrainian protest and investigate the role of Canadian government in the protest.⏎3 Views 2 Downloads<|endoftext|>The environment contains those factors ‘surrounding the organization’.

  4. #4 peak

    /Rosetta/MPS for OSIRIS Team MPS/UPD/LAM/IAA/SSO/INTA/UPM/DASP/IDAThe spacecraft will spend the next year or so collecting data and photographing the comet which

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. L10:1244 excites
  2. L9:25 weakly excites
  3. L11:2444 weakly suppresses
  4. L10:1516 weakly excites
  5. L10:1340 weakly excites
  6. L4:7 weakly excites
  7. L11:1675 weakly suppresses
  8. L2:1481 weakly excites
  9. L8:381 weakly excites
  10. L10:1731 weakly excites
  11. L9:605 weakly excites
  12. L3:633 weakly suppresses
  13. L6:1019 weakly suppresses
  14. L5:396 weakly excites
  15. L2:1188 weakly excites
  16. L10:494 weakly suppresses
  17. L6:216 weakly excites
  18. L8:1821 weakly suppresses
  19. L3:2026 weakly suppresses
  20. L9:1517 weakly suppresses
  21. L1:3027 weakly suppresses
  22. L1:2969 weakly suppresses
  23. L10:2201 weakly suppresses
  24. L3:2369 weakly excites

Feeds later layers

  1. L20:325 weakly suppresses
  2. L22:2309 weakly excites
  3. L15:134 weakly excites
  4. L23:628 weakly suppresses
  5. L15:1865 weakly excites
  6. L14:1634 weakly excites
  7. L13:2645 weakly suppresses
  8. L18:1681 weakly suppresses
  9. L13:1498 weakly excites end of a document boundary
  10. L15:580 weakly suppresses math and physics formulas
  11. L15:2242 weakly suppresses
  12. L21:2944 weakly excites
  13. L14:153 weakly excites
  14. L17:1352 weakly excites
  15. L22:2179 weakly excites
  16. L16:2094 weakly suppresses
  17. L25:1453 weakly excites
  18. L15:2118 weakly suppresses
  19. L21:1774 weakly excites
  20. L24:2887 weakly excites
  21. L21:2462 weakly suppresses
  22. L13:724 weakly excites
  23. L25:2568 weakly excites
  24. L23:1218 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 -2.69 and -2.551 values between -2.16 and -2.032 values between -1.90 and -1.772 values between -1.77 and -1.646 values between -1.64 and -1.515 values between -1.51 and -1.378 values between -1.37 and -1.2412 values between -1.24 and -1.1129 values between -1.11 and -0.9834 values between -0.98 and -0.8546 values between -0.85 and -0.72108 values between -0.72 and -0.59175 values between -0.59 and -0.46293 values between -0.46 and -0.32493 values between -0.32 and -0.19833 values between -0.19 and -0.061357 values between -0.06 and 0.07338 values between 0.07 and 0.2086 values between 0.20 and 0.3333 values between 0.33 and 0.4621 values between 0.46 and 0.597 values between 0.59 and 0.736 values between 0.73 and 0.861 values between 0.86 and 0.992 values between 0.99 and 1.123 values between 1.12 and 1.251 values between 1.25 and 1.381 values between 1.51 and 1.641 values between 1.64 and 1.781 values between 2.04 and 2.171 values between 2.30 and 2.431 values between 2.43 and 2.56offnegativepositiveoften

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. 什么样的
  3. 급
  4. 图
  5. اض
  6. 或者
  7. 说实话
  8. 也是非常
  9. 的具体
  10. eder

Suppresses

  1. Constructed
  2. 相助
  3. eness
  4. addElement
  5. udas
  6. castle
  7. rieve
  8. =all
  9. okin
  10. 栊