Neuron Explorer

Qwen3-0.6B

pipe before a table field label

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 reacts to the vertical bar that starts a new row in a table of facts, such as the bar just before 'Number of pages' or 'Owner'. When it was turned up, the model's writing did not change in any clear way that matched this pattern.

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

    portal/species/overview?p_p_spp=257336⏎This map is based on the

  2. #2 peak

    |The Physical Object|⏎|Number of Pages||136|⏎Learn why the Common Core is important for

  3. #3 peak

    Place of Publication||Perth, Western Australia|⏎|Number of pages||173|⏎|ISBN (Print)||9 7

  4. #4 peak

    flag staff|⏎|- elevation||15.75 ft (4.8 m) over stone base|⏎July 3, 1929⏎|Owner||National Park Service|⏎|Easiest access||walkway from avenue

  5. #5 peak

    you will need the Adobe Digital Editions app, or BlueFire Reader or Txtr app. These are free, too. For more details see this article.⏎|Size: ||359 KB|⏎|Date published: ||

  6. #6 peak

    an exception, which is life as usual for a Python programmer.⏎if not server_response['successful']:⏎etag = server_response['file_handle'].info().getheader('ETag')⏎xml_document = server_response['xml_document']<|endoftext|>Deepak V

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... I have been thinking... I have been thinking... I have been thinking... I have been thinking... and I have been thinking... I have been thinking... and I have been thinking... I have been thinking...

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

    0 00:10:22 CEST 2010⏎On 9/9/10 4:39 PM, Baba wrote:⏎> In below code "

  2. #2 peak

    December 6, 2010 %9 text/html %U http://blogs.discovermagazine.com/badastronomy/2008/10/13/death-by-meteorite/⏎Disclaimer: ComPAD

  3. #3 peak

    ' Classroom.|⏎|Levels:||Late Elem. (3-5), Middle School (6-8), High

  4. #4 peak

    and then its right child.⏎T / \ I S / \ D E<|endoftext|>Original URL: http://www.theregister.co.uk/2006/08/21/mars_geysers/⏎Martian pole freck

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. L11:2834 suppresses
  2. L4:536 suppresses
  3. L4:19 weakly suppresses
  4. L6:398 weakly excites
  5. L8:2316 weakly suppresses
  6. L2:2791 weakly suppresses
  7. L11:2352 weakly suppresses
  8. L10:720 weakly suppresses
  9. L9:2433 weakly excites
  10. L2:1055 weakly excites
  11. L11:1174 weakly excites
  12. L1:2934 weakly excites
  13. L4:1040 weakly excites
  14. L11:253 weakly excites
  15. L8:193 weakly excites
  16. L7:1860 weakly suppresses
  17. L11:2089 weakly suppresses
  18. L11:515 weakly excites 'hands' in hand-washing advice
  19. L11:417 weakly excites
  20. L9:120 weakly excites
  21. L3:2135 weakly suppresses
  22. L5:1594 weakly excites
  23. L4:2992 weakly excites
  24. L7:315 weakly excites

Feeds later layers

  1. L16:1124 excites
  2. L13:2141 excites end-of-text token after finished articles
  3. L16:1913 excites
  4. L20:904 excites
  5. L19:2132 excites
  6. L19:1717 excites
  7. L13:901 excites
  8. L20:941 excites
  9. L14:1517 excites
  10. L15:2090 excites
  11. L16:2988 excites
  12. L16:1790 excites
  13. L21:738 weakly excites
  14. L19:870 weakly suppresses
  15. L17:882 weakly excites
  16. L13:625 weakly excites
  17. L18:1502 weakly excites
  18. L17:305 weakly excites
  19. L18:2204 weakly excites
  20. L17:2079 weakly suppresses
  21. L22:319 weakly excites
  22. L18:557 weakly excites
  23. L16:1447 weakly excites
  24. L15:1673 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.

2 values between -0.62 and -0.576 values between -0.57 and -0.525 values between -0.52 and -0.4714 values between -0.47 and -0.4231 values between -0.42 and -0.3752 values between -0.37 and -0.3291 values between -0.32 and -0.28146 values between -0.28 and -0.23237 values between -0.23 and -0.18320 values between -0.18 and -0.13418 values between -0.13 and -0.08487 values between -0.08 and -0.03592 values between -0.03 and 0.02454 values between 0.02 and 0.07402 values between 0.07 and 0.12280 values between 0.12 and 0.17180 values between 0.17 and 0.2299 values between 0.22 and 0.2747 values between 0.27 and 0.3225 values between 0.32 and 0.3712 values between 0.37 and 0.425 values between 0.42 and 0.462 values between 0.51 and 0.561 values between 1.30 and 1.35offnegativepositiveoften

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. ewidth
  2. anggal
  3. scope
  4. rieg
  5. 啥
  6. scope
  7. AÇÃO
  8. lement
  9. lijk
  10. göre

Suppresses

  1. 串
  2. 就够了
  3. 镑
  4. .getOutputStream
  5. öst
  6. othy
  7. compromises
  8. ((__
  9. ören
  10. ByteArrayOutputStream