Neuron Explorer

Qwen3-0.6B

'key' after 'The'

A neuron in layer 0 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 fires very strongly on the word 'key' when it follows 'The', as in 'The key to' or 'The key is'. It is slightly negative on 'of' after 'interior'. Turning it up or down did not make the model's writing talk about keys or important points.

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

    The key here is precisely the geography that cannot be seen from above: the underwater domain where

  2. #2 peak

    interiors even when the windows are closed. Designs often incorporate atria or internal stairwells to take advantage of these effects.⏎Winter environmental ventilation⏎The key issue for winter environmental ventilation is the control of indoor air quality. Experience shows that

  3. #3 peak

    or reply design paradigm for synchronous and asynchronous applications. An application’s business logic or individual functions are modularized and presented as services for consumer or client applications. The key to these services is their loosely coupled nature; i.e., the service interface is

  4. #4 peak

    ⏎- ‘The key to making the discovery was the development of new techniques that allow monosynaptic

  5. #5 peak

    but keeps coming back.” Or “I took (said extended period of time off) and it still didn’t help and I have the same pain!”⏎The key component that is missing from most therapy programs is the strengthening aspect. Yes, the

  6. #6 peak

    working with a personal trainer.⏎This guide’s main goal is to help you become more active, get the most from your activities, and do them safely. The key is to know your starting point and build slowly from there. Knowing your starting point

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 the universe. It's a place where the stars are dancing in the sky. It's a place where time flies in the air. It's a place where everything is possible. It's a place where the universe is 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

    of Liberty.The interior of the structure is a collection of wrought iron rivets, saddles and armature

  2. #2 peak

    varying shades of pink, yellow and creamy white. The interior of the lip is richly feathered, and is dark crimson. The rich contrast

  3. #3 peak

    . While Grant set out to destroy Lee’s army, Sherman was ordered to move against Johnston’s army and also to do as much damage as possible to the interior of Georgia.⏎Sherman interpreted his orders as allowing him to concentrate on the capture of

  4. #4 peak

    The completed rest house on the opening day⏎A typical bedroom in the rest house⏎Inside

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

None: this is the first layer.

Feeds later layers

  1. L2:330 excites
  2. L3:774 excites
  3. L19:1019 weakly suppresses
  4. L1:176 weakly excites
  5. L5:613 weakly excites
  6. L22:848 weakly excites
  7. L1:697 weakly excites
  8. L18:1580 weakly suppresses
  9. L19:2444 weakly suppresses
  10. L1:233 weakly excites
  11. L19:1645 weakly suppresses
  12. L6:604 weakly excites
  13. L14:866 weakly suppresses
  14. L10:315 weakly suppresses
  15. L2:91 weakly suppresses
  16. L1:2209 weakly suppresses
  17. L17:1967 weakly excites
  18. L22:770 weakly suppresses
  19. L12:665 weakly suppresses
  20. L20:987 weakly suppresses
  21. L9:837 weakly suppresses
  22. L16:151 weakly excites
  23. L24:1137 weakly excites
  24. L17:359 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.

1 values between -0.94 and -0.861 values between -0.77 and -0.681 values between -0.68 and -0.593 values between -0.59 and -0.504 values between -0.50 and -0.419 values between -0.41 and -0.3220 values between -0.32 and -0.2452 values between -0.24 and -0.15317 values between -0.15 and -0.062684 values between -0.06 and 0.03659 values between 0.03 and 0.12101 values between 0.12 and 0.2126 values between 0.21 and 0.303 values between 0.30 and 0.397 values between 0.39 and 0.473 values between 0.47 and 0.562 values between 0.56 and 0.655 values between 0.65 and 0.742 values between 0.74 and 0.832 values between 0.83 and 0.921 values between 0.92 and 1.001 values between 1.09 and 1.181 values between 1.18 and 1.271 values between 1.36 and 1.451 values between 2.42 and 2.511 values between 2.51 and 2.60offnegativepositiveoften

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. وس
  2. ought
  3. Fabric
  4. renc
  5. React
  6. igli
  7. ranch
  8. ront
  9. 秘密
  10. 养老服务

Suppresses

  1. containing
  2. powered
  3. bourg
  4. Uhr
  5. currentItem
  6. 全套
  7. humanities
  8. бур
  9. огда
  10. courier