Neuron Explorer

Qwen3-0.6B

document boundary end-of-text token

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 responds to the marker that signals one web page or article has ended and an unrelated one begins. Turning it up produced rambling, unrelated text, so the steering does not show a clear meaning.

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

    and preventing oral diseases before they become more serious. Supreme Dental CT helps you to keep a beautiful smile and improve your overall health by making these practices a habit.<|endoftext|>Teen Attitude: What is normal? How does it affect behaviour?⏎Posted on

  2. #2 peak

    and tissue paper onto our tins too. The flower pots are looking fantastic. We can't wait to fill them with soil and seeds once they are dry.<|endoftext|>NASA's Plan for Real-Life Stasis Could Get Us to Mars⏎Countless

  3. #3 peak

    ium Pro and the complexity of the architecture, it's surprising that this chip runs as fast as it does. Yet, performance is not a problem--quite the opposite, actually. Using BAPCo benchmarks, the 150-MHz

  4. #4 peak

    pro quo. We watch legislators and other elected officials make off with our hard-earned money, and we get nothing in return.⏎Joe Bruno can now rest in peace. It’s been nine years since he was charged with accepting bribes from a

  5. #5 peak

    ”). This is good for the government because it is lower than what we believe to be the actual price inflation rate and makes GDP look better than it really is.<|endoftext|>For the genre including games like Final Fantasy and Summon Night:

  6. #6 peak

    to provide service. Aaron lives in beautiful Vancouver but spends lots of time in the Sea to Sky corridor including Whistler, Pemberton and Squamish.<|endoftext|>The current heatwave in northern Europe was made twice as likely by climate change,

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: if you want to build a more advanced, more interactive, and more intelligent, more intelligent and interactive, and more intelligent and collaborative, then that's the result of the future where the world is built on the idea of the future with a

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

    with coagulopathy have a threefold to fourfold greater mortality, and are up to eight times more likely to die within the first 24 hours following injury [2-4,7,8].⏎Injury

  2. #2 peak

    221293000 people.⏎Pakistan telephone code is 92.<|endoftext|>T HERE are as many definitions of public opinion as there aredisciplinary points of view. Political scientists see public opinion interms of the distribution of

  3. #3 peak

    , a tree can settle in too deeply. Amending soil beneath the root ball will not encourage roots. Most of a tree's roots will develop in the top 12 to 18 inches of soil, although the depth depends on a

  4. #4 peak

    advised. There are thermometers available at reasonable price which can help you determine the temperature. A report suggests that keeping the fridge cool at 5 degrees C or below can prevent germs from growing and multiplying.⏎2. Ensure Food is stored at

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. L7:211 suppresses
  2. L12:2283 suppresses
  3. L6:12 suppresses
  4. L9:2266 weakly suppresses
  5. L6:18 weakly suppresses
  6. L6:576 weakly suppresses
  7. L12:2060 weakly excites
  8. L10:49 weakly suppresses
  9. L1:2830 weakly suppresses
  10. L4:953 weakly suppresses
  11. L8:7 weakly suppresses
  12. L8:3 weakly suppresses
  13. L2:733 weakly suppresses
  14. L7:122 weakly suppresses
  15. L12:1463 weakly excites
  16. L9:1624 weakly suppresses
  17. L3:1631 weakly excites
  18. L7:1082 weakly excites
  19. L12:1573 weakly excites
  20. L7:186 weakly excites
  21. L11:2839 weakly excites
  22. L3:3066 weakly excites
  23. L12:739 weakly excites
  24. L6:2572 weakly excites

Feeds later layers

  1. L17:2779 suppresses
  2. L16:1496 suppresses
  3. L15:515 suppresses word before 'that', 'to' or 'for' continuation
  4. L15:2618 suppresses
  5. L16:278 suppresses
  6. L14:1378 excites
  7. L15:1239 suppresses
  8. L14:1568 excites
  9. L16:1271 suppresses
  10. L15:1326 weakly excites
  11. L14:1993 weakly excites
  12. L21:2920 weakly excites
  13. L15:1634 weakly excites
  14. L15:2519 weakly excites
  15. L16:1267 weakly suppresses
  16. L15:2491 weakly excites
  17. L18:2644 weakly excites
  18. L17:1399 weakly excites
  19. L14:709 weakly excites
  20. L15:2931 weakly excites
  21. L14:96 weakly suppresses
  22. L17:980 weakly suppresses
  23. L19:2574 weakly excites
  24. L16:2673 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 -2.49 and -2.351 values between -2.35 and -2.214 values between -2.21 and -2.076 values between -1.94 and -1.804 values between -1.80 and -1.665 values between -1.66 and -1.528 values between -1.52 and -1.3914 values between -1.39 and -1.2535 values between -1.25 and -1.1146 values between -1.11 and -0.9761 values between -0.97 and -0.84105 values between -0.84 and -0.70138 values between -0.70 and -0.56192 values between -0.56 and -0.42286 values between -0.42 and -0.29392 values between -0.29 and -0.15439 values between -0.15 and -0.01518 values between -0.01 and 0.13467 values between 0.13 and 0.27400 values between 0.27 and 0.40289 values between 0.40 and 0.54186 values between 0.54 and 0.68107 values between 0.68 and 0.8274 values between 0.82 and 0.9552 values between 0.95 and 1.0925 values between 1.09 and 1.2321 values between 1.23 and 1.3710 values between 1.37 and 1.507 values between 1.50 and 1.646 values between 1.64 and 1.783 values between 1.78 and 1.922 values between 1.92 and 2.051 values between 2.05 and 2.191 values between 2.19 and 2.331 values between 2.88 and 3.02offnegativepositiveoften

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. 他还
  2. utsche
  3. ","#
  4. {}).
  5. "]=="
  6. rift
  7. eacher
  8. >All
  9. ('?
  10. :%

Suppresses

  1. concentrates
  2. useful
  3. helpful
  4. Helpful
  5. opsis
  6. arg
  7. insi
  8. 腿部
  9. con
  10. правля