Neuron Explorer

Qwen3-0.6B

unclear: link fragments and document ends

A neuron in layer 13 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 is hard to pin down. Its strongest moments come inside web links, such as the middle of a YouTube or Twitter address, and right where one web page ends and another begins. The steering outputs do not show a clear link to these spots.

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

    et paper. At a much less sophisticated level, PutnamPig’s “Swine Flu Song” on Youtube http://www.youtube.com/watch?v=tbt_PuVAVTU defends pigs against unfair stigmatisation over H1N1

  2. #2 peak

    8, July 6). 1929 Stock Market Crash and the Great Depression –⏎Documentary [Video File]. Retrieved from https://youtu.be/qlSxPouPCIM⏎The post 1929 Stock

  3. #3 peak

    To give a sense of the true dimensions of the Larsen C iceberg, here's an animation based on real data, with something familiar for scale pic.twitter.com/OdQA07qhBb⏎— Adrian Luckman (@adrian_l

  4. #4 peak

    I highly recommend reading Bruce Rogers' post:⏎Good study for the Moxy people. I did a post and review on that article. https://t.co/D9NIorgXuc— Bruce Rogers (@bjrmd) August 30

  5. #5 peak

    geographic regions.(opens in new tab)⏎Journal of Clinical Microbiology⏎Suleyman FelekY Rikihisa⏎George RidingPeter Willadsen<|endoftext|>Today, various types of compressed gases such as acetylene, propane, oxygen

  6. #6 peak

    014⏎Originally published: August 18, 2014. Last Updated: August 18, 2014.<|endoftext|>The more I read about this, the more confused I become. Illuminance,

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 a 3D space, and the next step is to go to the next step, and the next step is to go to 4th step, and the next step is to go to 5th step, and 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

    ) hypotonic D) none of these I think it's B. I'm not too sure though.⏎November 27, 2007<|endoftext|>Hopefully your child will not experience at first hand the impact of war. But sadly

  2. #2 peak

    oviridae family. …⏎- Noroviruses. …⏎- Pasteurella. …⏎- Yersinia enterocolitica.<|endoftext|>LACMA May Fund The Fading Watts Towers⏎Due to a severe budget

  3. #3 peak

    ycetica<|endoftext|>Adding Fractions Level 3⏎The 5-step planIn Adding Fractions

  4. #4 peak

    99) 23: 25. doi:10.1023/A:1005462530658⏎- 896 Downloads⏎Using recently available data from China

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. L10:2913 suppresses
  2. L5:3040 weakly suppresses
  3. L9:1028 weakly excites
  4. L10:57 weakly suppresses
  5. L6:2839 weakly suppresses
  6. L12:1721 weakly excites
  7. L6:2556 weakly suppresses
  8. L12:1877 weakly suppresses
  9. L10:2001 weakly excites
  10. L10:2621 weakly excites
  11. L2:2740 weakly suppresses 'on' before time or track
  12. L8:10 weakly excites
  13. L8:1837 weakly suppresses
  14. L1:51 weakly excites
  15. L4:2760 weakly excites
  16. L12:1664 weakly excites end of document token
  17. L12:59 weakly suppresses
  18. L3:3011 weakly suppresses
  19. L5:198 weakly excites unclear: weak "and" in lists
  20. L12:2472 weakly excites
  21. L11:1539 weakly excites
  22. L6:214 weakly suppresses
  23. L6:1710 weakly excites
  24. L8:2395 weakly suppresses

Feeds later layers

  1. L19:738 excites
  2. L18:921 weakly excites
  3. L25:1816 weakly excites
  4. L17:1571 weakly suppresses
  5. L14:2391 weakly suppresses
  6. L16:1451 weakly suppresses
  7. L16:1161 weakly suppresses
  8. L27:877 weakly excites
  9. L17:1175 weakly suppresses
  10. L14:471 weakly suppresses
  11. L18:1723 weakly excites
  12. L20:2754 weakly suppresses
  13. L26:135 weakly excites
  14. L24:782 weakly excites
  15. L21:2125 weakly suppresses
  16. L24:483 weakly suppresses
  17. L17:1252 weakly suppresses
  18. L14:118 weakly suppresses
  19. L16:2904 weakly excites
  20. L25:2846 weakly excites
  21. L27:1023 weakly excites
  22. L14:2291 weakly suppresses
  23. L18:2002 weakly suppresses
  24. L18:672 weakly excites zero right after a decimal point

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 -1.15 and -0.9033 values between -0.90 and -0.65193 values between -0.65 and -0.411220 values between -0.41 and -0.162248 values between -0.16 and 0.09198 values between 0.09 and 0.345 values between 0.34 and 0.591 values between 1.59 and 1.841 values between 2.84 and 3.091 values between 4.09 and 4.342 values between 5.09 and 5.341 values between 6.58 and 6.832 values between 7.33 and 7.581 values between 8.58 and 8.83offnegativepositiveoften

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. 钋
  2. aye
  3. illas
  4. weg
  5. 体制
  6. ás
  7. encoding
  8. 世家
  9. dq
  10. qw

Suppresses

  1. �
  2. iterate
  3. leting
  4. $is
  5. rement
  6. оч
  7. ﷺ
  8. ometric
  9. -second
  10. ходит