Neuron Explorer

Qwen3-0.6B

end-of-document boundary tokens

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 hidden marker that signals the end of one web page and the start of another. It is most active on the last word just before that break. Turning it up made the writing wander and repeat, with no clear topic.

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

    body structure while another one is about body function. But if you want to study them briefly and combine then you can download human anatomy and physiology at the end.<|endoftext|>Increasing the impact of research via wider dissemination of the findings is a well-known strategy

  2. #2 peak

    Exposure Attention Interpretation Exposure Exposure highlights the stimulus to which a consumer is exposed to. Exposure can be either deliberate or random. ...⏎Not exactly what you need?<|endoftext|>Allegany County’s

  3. #3 peak

    of the stock has dropped due to poor fundamentals. There are a few ways to check if a dividend is safe, and this post will teach you just that.<|endoftext|>3 September 2016⏎Layer by layer, building with eroision

  4. #4 peak

    utf8mb4) bytes per character can be used. So if one specifies VARCHAR(50) CHARSET utf8mb4, …[Read more]<|endoftext|>To actually reduce pounds in a healthier manner, it is perfect to burn fat gradually

  5. #5 peak

    Environmental management Observe/Discuss:⏎- Environmental management Let’s Discuss:⏎- Environmental management Observe:⏎- Environmental management Let’s think:⏎- Environmental management Paragraph<|endoftext|>Organic More

  6. #6 peak

    where certain mail servers is given as authorized mail server for your domain. The recipient after receiving the mail from your domain, can verify with your domain SPF record to<|endoftext|>Researchers from the Sloan Digital Sky Survey (SDSS-II) announced the discovery of

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 will not be able to stop myself from being... I have been thinking about it... and I will not be able to stop myself from... I will be the same. Okay, that's all. I have been thinking about it...

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

    21, 1 PM⏎World StampsNov 22, 2021, 7 PM<|endoftext|>At SACRE, the Fluent CFD code has been used in the past 15 years for research as well as

  2. #2 peak

    able to develop more focused methods to control hESC differentiation and move closer to clinical therapies. (ANI)<|endoftext|>Blaeu, Joan (1596-1673)⏎[Original engraving and etching cartography/

  3. #3 peak

    the 1890s the Night Riders had done their part, Jim Crow laws were firmly in place and the Confederate Party had a strong power base in the

  4. #4 peak

    between the voltage and current, it is known as the active power. The second term is time-varying, it depends<|endoftext|>The Difference Between LDL and HDL Cholesterol: Blood cholesterol is an important indicator of cardiovascular health. Maintaining optimal

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. L8:108 suppresses
  2. L9:1399 excites
  3. L8:790 excites
  4. L8:436 suppresses
  5. L8:163 suppresses
  6. L9:1148 suppresses
  7. L11:634 suppresses
  8. L5:76 suppresses
  9. L8:528 excites
  10. L11:1150 excites
  11. L6:123 suppresses
  12. L8:1119 weakly suppresses
  13. L11:835 weakly suppresses
  14. L9:2258 weakly suppresses
  15. L11:1263 weakly suppresses
  16. L9:1131 weakly suppresses
  17. L10:530 weakly suppresses
  18. L8:824 weakly suppresses
  19. L11:1379 weakly excites
  20. L5:159 weakly suppresses
  21. L11:452 weakly suppresses
  22. L2:566 weakly suppresses
  23. L8:80 weakly suppresses
  24. L2:637 weakly suppresses

Feeds later layers

  1. L14:2682 suppresses
  2. L16:787 suppresses
  3. L13:1471 excites
  4. L14:2600 excites
  5. L18:1288 suppresses
  6. L14:2198 excites
  7. L13:853 excites
  8. L13:397 suppresses
  9. L17:574 suppresses
  10. L13:2371 suppresses
  11. L13:2774 excites
  12. L14:2525 suppresses
  13. L14:2032 suppresses
  14. L13:1612 weakly suppresses
  15. L13:1236 weakly suppresses
  16. L14:1023 weakly suppresses
  17. L13:2037 weakly suppresses
  18. L13:201 weakly suppresses
  19. L15:1346 weakly excites
  20. L14:2574 weakly suppresses
  21. L13:2336 weakly suppresses
  22. L15:1121 weakly suppresses
  23. L13:1119 weakly suppresses
  24. L19:1193 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.

2 values between -3.20 and -3.053 values between -2.31 and -2.172 values between -2.17 and -2.021 values between -2.02 and -1.873 values between -1.87 and -1.732 values between -1.73 and -1.585 values between -1.58 and -1.435 values between -1.43 and -1.298 values between -1.29 and -1.1414 values between -1.14 and -0.9918 values between -0.99 and -0.8534 values between -0.85 and -0.7046 values between -0.70 and -0.5555 values between -0.55 and -0.40140 values between -0.40 and -0.26467 values between -0.26 and -0.111944 values between -0.11 and 0.04724 values between 0.04 and 0.18223 values between 0.18 and 0.3392 values between 0.33 and 0.4840 values between 0.48 and 0.6322 values between 0.63 and 0.7723 values between 0.77 and 0.9212 values between 0.92 and 1.075 values between 1.07 and 1.214 values between 1.21 and 1.362 values between 1.36 and 1.515 values between 1.51 and 1.652 values between 1.65 and 1.801 values between 1.80 and 1.951 values between 1.95 and 2.092 values between 2.09 and 2.241 values between 2.54 and 2.68offnegativepositiveoften

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. Ping
  2. EAR
  3. lyn
  4. awan
  5. ude
  6. ihat
  7. ilon
  8. úb
  9. Faces
  10. ($__

Suppresses

  1. ptoms
  2. Автор
  3. rosso
  4. 有效性
  5. 整车
  6. ��
  7. ERCHANT
  8. мом
  9. QUIRED
  10. ОР