Neuron Explorer

Qwen3-0.6B

the word role after play or a potential role

A neuron in layer 6 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the words role and a in phrases like play a role or the role of something. Turning it up did not produce any clear change in topic, so its job is hard to pin down from the outputs.

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

    If you don’t see the Drawing Tool Shapes stencil, add it to your Shapes window. Click More Shapes > Visio Extras > Drawing Tool Shapes.⏎From Drawing Tool Shapes, drag the Measure tool shape onto the drawing page.⏎Glue one endpoint

  2. #2 peak

    injury or inflammation during pregnancy – could steer sexual behavior development in offspring. It’s even conceivable that taking antihistamines or pain relievers during pregnancy could play a role.⏎“We’re really interested in the fundamental mechanisms that drive brain development and sex

  3. #3 peak

    flutter" and "Control of ventricular rate in atrial flutter" and "Atrial flutter: Maintenance of sinus rhythm" and "Embolic risk and the role of anticoagulation in

  4. #4 peak

    to genetics, there is evidence that AD has a microbial component. More specifically, research has linked the disease to Staphylococcus bacteria colonization that may play a role in flares and disease control. The bacteria and human genetics are thought to

  5. #5 peak

    educational access in rural areas. However, it is one part of the solution with multiple and diverse challenges faced by rural schools.⏎- Off-grid renewable energies can play a key role in school electrification.<|endoftext|>Lazy evaluation means that expressions are not evaluated

  6. #6 peak

    with when athletes return to action after concussions, according to a new study of collegiate athletes and recovery time. “Although preliminary, the current results highlight the potential role of biomarkers in tracking neuronal recovery, which may be associated with duration 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 have a 3D printed model. I can't get it right. The model is in a box with a 3D printer. I can't get it right. I have a 3D model that is in a box

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

    American Migration Experience, the Schomburg Center for Research in Black Culture.⏎- Chicago and the Great Migration, The Newberry.⏎- "Sir I Will Thank You with All My Heart": Seven Letters from the Great Migration, History Matters.<|endoftext|>

  2. #2 peak

    screams’ her outrage at the limitations for women in Emma.⏎When accosted by Mrs. Elton, Jane Fairfax says,⏎“Excuse me, ma’am, but this is by no means my intention; I make no inquiry

  3. #3 peak

    Let’s see if you know any of these enormous structures.⏎Let’s check out these huge black holes in the

  4. #4 peak

    others would appreciate reading this, please share it through the social media buttons.⏎When Mary arrived and saw Jesus, she fell at his feet and said, “Lord, if only you had been here, my brother would not have died.”⏎“Where

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. L4:63 excites
  2. L3:445 suppresses
  3. L3:694 suppresses
  4. L4:454 excites
  5. L5:371 weakly excites
  6. L5:167 weakly suppresses
  7. L5:592 weakly suppresses
  8. L5:151 weakly suppresses
  9. L5:387 weakly excites
  10. L3:125 weakly suppresses verb right after a subject in clauses
  11. L5:1466 weakly excites
  12. L5:397 weakly suppresses
  13. L5:1800 weakly suppresses
  14. L2:861 weakly excites
  15. L4:1615 weakly excites
  16. L2:531 weakly excites
  17. L4:150 weakly excites
  18. L5:2019 weakly suppresses
  19. L3:147 weakly excites
  20. L4:1718 weakly suppresses
  21. L5:771 weakly excites
  22. L3:987 weakly suppresses
  23. L4:1175 weakly excites
  24. L3:776 weakly excites

Feeds later layers

  1. L7:314 strongly excites
  2. L8:584 excites
  3. L15:720 excites
  4. L8:1994 excites
  5. L9:321 excites
  6. L9:778 excites
  7. L13:1853 excites
  8. L11:1286 excites
  9. L11:939 excites
  10. L10:1783 suppresses
  11. L8:355 excites
  12. L7:182 excites
  13. L7:656 excites
  14. L8:606 excites
  15. L14:2077 excites
  16. L11:923 excites
  17. L7:658 excites
  18. L12:1288 excites
  19. L18:1550 excites
  20. L13:507 excites
  21. L7:1569 excites
  22. L13:1339 suppresses
  23. L8:406 excites word after a question starter
  24. L7:1562 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.

1 values between -2.69 and -2.611 values between -2.18 and -2.102 values between -1.76 and -1.672 values between -1.67 and -1.593 values between -1.42 and -1.335 values between -1.16 and -1.081 values between -1.08 and -0.992 values between -0.99 and -0.912 values between -0.91 and -0.826 values between -0.82 and -0.744 values between -0.74 and -0.655 values between -0.65 and -0.5715 values between -0.57 and -0.4813 values between -0.48 and -0.4039 values between -0.40 and -0.3161 values between -0.31 and -0.23123 values between -0.23 and -0.14286 values between -0.14 and -0.06968 values between -0.06 and 0.03956 values between 0.03 and 0.12756 values between 0.12 and 0.20413 values between 0.20 and 0.29157 values between 0.29 and 0.3758 values between 0.37 and 0.4622 values between 0.46 and 0.544 values between 0.54 and 0.633 values between 0.63 and 0.71offnegativepositiveoften

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. мет
  2. venida
  3. 哪一个
  4. AtPath
  5. igator
  6. 任何时候
  7. itemType
  8. казино
  9. !important
  10. 餐桌

Suppresses

  1. ierce
  2. esign
  3. intensely
  4. iously
  5. apologized
  6. 용
  7. proceeded
  8. 日正式
  9. uses
  10. intens