Neuron Explorer

Qwen3-0.6B

verb after a gerund subject

A neuron in layer 11 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

The neuron fires on the verb 'is', 'has' or 'be' that comes right after a subject made of an activity word, such as 'Moderating your food consumption [is]' or 'Blaming the victim [is]'. Turning it up made the model's output drift into repetitive, unrelated sentences, so the steering does not confirm 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

    , sharing a common⏎German culture and language.⏎In Europe that scarcely mattered and had not for a century or more.⏎Determined<|endoftext|>Moderating your food consumption is the most important component of a healthy, balanced diet. Incorporating concepts such as

  2. #2 peak

    cognitive processes and to support brain function.⏎Almond also works wonders for heart health. Magnesium helps reduce the risk of cardiac arrest. Improving circulation of blood has been added to one of their added benefits. A diet containing with this as a

  3. #3 peak

    to staying up-to-date, boosting your immune system is wellness 101. Better yet, it’s simpler than you might think

  4. #4 peak

    other area of life we see something but it takes a leap of faith to declare it is real.⏎When you look at<|endoftext|>Blaming the victim is a practice that is exhibited within society as a means of justifying racism, sexism

  5. #5 peak

    ifier. The company financed design and implementation work from its own funds, bearing the risk associated with the investment.⏎Designing an efficient and reliable gasifier proved to be the main difficulty. Design and preparatory work lasted a year. Launching this

  6. #6 peak

    with a digital electronic camera and after that publish them for sale or displaying in a gallery.⏎Publish digital photography is one more late 1800s creation that redefined the guidelines for

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 number of 2000, the number of 2000, the number of 2000, the number of 2000, the number of 2000, the number of

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

    the shade are preferred. Females lay their eggs singly on the sides of the container at the water line in batches of 30-50. Egg laying occurs over a period of several days.⏎Exposure to high humidity at the water

  2. #2 peak

    that have already solidified. Only this way can the current and previous layers chemically bond and become a single solid object.⏎This phenomenon is known as overcuring. For the reasons just described, this overcuring is necessary and cannot be

  3. #3 peak

    Adafruit strives to be a fair, informative, and positive voice within the community – check it out here: adafruit.com/editorialstandards⏎Stop breadboarding and soldering – start making immediately! Adafruit’s Circuit Playground is jam-packed

  4. #4 peak

    be published, broadcast, rewritten or redistributed<|endoftext|>Wellness Connection Newsletter⏎Hello community members big and small!⏎The mind-body

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. L6:350 strongly excites
  2. L9:413 excites
  3. L2:2152 suppresses
  4. L9:527 suppresses
  5. L10:388 suppresses
  6. L8:234 suppresses
  7. L7:984 suppresses
  8. L7:418 excites
  9. L9:1240 excites
  10. L8:345 excites
  11. L6:318 suppresses
  12. L4:2803 suppresses
  13. L7:1243 excites
  14. L10:2323 excites
  15. L6:2322 excites
  16. L9:388 excites
  17. L6:290 suppresses
  18. L3:1204 excites
  19. L3:110 excites
  20. L9:944 suppresses
  21. L10:2811 excites
  22. L9:2200 suppresses
  23. L5:2680 suppresses
  24. L4:253 suppresses

Feeds later layers

  1. L16:2826 excites
  2. L16:1641 excites
  3. L20:1501 excites
  4. L18:353 excites
  5. L19:1307 excites
  6. L19:2163 excites
  7. L16:2288 excites
  8. L18:1793 excites
  9. L12:672 excites
  10. L12:1424 excites
  11. L15:800 suppresses
  12. L12:1890 excites
  13. L21:781 excites
  14. L13:612 excites
  15. L12:2287 excites
  16. L12:1032 suppresses
  17. L16:981 excites
  18. L16:1522 excites
  19. L12:1342 excites
  20. L14:1585 excites
  21. L15:1931 excites
  22. L17:1440 excites
  23. L18:2725 excites
  24. L12:1196 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.

1 values between -1.67 and -1.561 values between -1.24 and -1.141 values between -0.93 and -0.824 values between -0.82 and -0.7217 values between -0.72 and -0.6181 values between -0.61 and -0.51270 values between -0.51 and -0.40611 values between -0.40 and -0.30954 values between -0.30 and -0.19934 values between -0.19 and -0.09603 values between -0.09 and 0.02191 values between 0.02 and 0.1369 values between 0.13 and 0.2334 values between 0.23 and 0.3429 values between 0.34 and 0.4431 values between 0.44 and 0.5516 values between 0.55 and 0.6512 values between 0.65 and 0.7611 values between 0.76 and 0.865 values between 0.86 and 0.976 values between 0.97 and 1.075 values between 1.07 and 1.182 values between 1.18 and 1.282 values between 1.28 and 1.391 values between 1.39 and 1.493 values between 1.49 and 1.602 values between 1.60 and 1.702 values between 1.70 and 1.815 values between 1.81 and 1.923 values between 1.92 and 2.021 values between 2.02 and 2.131 values between 2.44 and 2.55offnegativepositiveoften

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. 的过程
  2. 过程
  3. 的方式
  4. 办
  5. done
  6. 的过程中
  7. Done
  8. process
  9. 过程中
  10. Ę

Suppresses

  1. inke
  2. leston
  3. 生命力
  4. arel
  5. anch
  6. astle
  7. oque
  8. osity
  9. lename
  10. 利用率