Neuron Explorer

Qwen3-0.6B

unclear: 'thatch' and 'the following'

A neuron in layer 1 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 a few unrelated word pieces, such as the 'ch' in 'thatch' and the word 'following' after 'The'. It also pushes strongly negative on 'of' in phrases like 'this type of'. There is no single clear theme, and turning it up did not reveal one.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    relationships, mutual care and responsibility.⏎Relational theologies say something really interesting: This how we’re wired, because it’s how God is wired.⏎Here’s what they

  2. #2 peak

    ens the grass, fosters thatch and encourages disease.⏎How to Have a Healthy Lawn and Yard⏎Fortunately, you

  3. #3 peak

    compacted soil by removing small plugs from the topmost layer of earth. Aeration should be considered for all high-traffic areas, lawns with excess thatch, and clay soil.⏎Reasons to Aerate your Lawn⏎Compact soil features

  4. #4 peak

    A certain amount of thatch is good for your lawn and will help reduce weeds and lock in moisture. Generally, a 1/2″ layer of thatch is acceptable but more indicates an excessive buildup. In addition to bagging lawn cl

  5. #5 peak

    entering the ground. Grass with shallow root systems have restricted access to water and nutrients, and result in poor draught tolerance. Additionally, compact soil causes excess thatch buildup as soil microorganisms are less effective. All these characteristics make for a weak

  6. #6 peak

    ," senior U.S. officials took it for granted that a massive Warsaw Pact conventional attack on Western Europe would prompt a nuclear response from outnumbered Western forces.⏎The following documents, a sampling from the 1963-64 period,

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 am going to be going to the other. I have been going to the other. I have been going to the other. I have been going to the other. I am going to be going to the other. I am going to

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

    A single-layer PCB is a device that has a copper layer on one side of the board. This type of PCB has components on one side and the pattern to conduct electricity on the other side

  2. #2 peak

    . One example of an informal team would be a group of co-workers who meet informally over lunch breaks to discuss problems with the work environment. Although this type of teamwork is formed without a formal structure, it can be extremely effective when team members

  3. #3 peak

    ays for something say, 6 months before you actually provide the service or product, how do you record such a payment? How do we account for this type of transaction? Enter the term "deferred revenue".

  4. #4 peak

    that are imported from the procurement supply warehouse or purchased locally. This type of warehouse is generally close to the commodity sales market, and the scale is generally smaller

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. L0:2182 weakly suppresses
  2. L0:592 weakly suppresses
  3. L0:170 weakly excites
  4. L0:973 weakly excites
  5. L0:787 weakly excites
  6. L0:865 weakly suppresses
  7. L0:1388 weakly suppresses
  8. L0:3016 weakly suppresses
  9. L0:502 weakly suppresses
  10. L0:352 weakly suppresses
  11. L0:1715 weakly suppresses
  12. L0:1336 weakly excites
  13. L0:567 weakly excites
  14. L0:1881 weakly suppresses
  15. L0:1522 weakly suppresses
  16. L0:946 weakly excites
  17. L0:2467 weakly excites
  18. L0:3071 weakly excites
  19. L0:535 weakly excites
  20. L0:918 weakly suppresses
  21. L0:1386 weakly suppresses
  22. L0:478 weakly suppresses
  23. L0:1111 weakly excites
  24. L0:428 weakly suppresses

Feeds later layers

  1. L16:2089 excites
  2. L4:270 weakly suppresses
  3. L12:521 weakly suppresses
  4. L7:1141 weakly excites
  5. L12:2781 weakly suppresses
  6. L13:612 weakly excites
  7. L20:1249 weakly excites
  8. L16:731 weakly excites
  9. L19:1788 weakly excites
  10. L4:1171 weakly excites
  11. L4:323 weakly suppresses
  12. L27:1387 weakly excites
  13. L10:1435 weakly excites
  14. L13:1744 weakly suppresses
  15. L6:398 weakly suppresses
  16. L4:326 weakly excites
  17. L9:1529 weakly excites
  18. L17:482 weakly excites
  19. L12:1811 weakly excites
  20. L17:1823 weakly excites
  21. L6:1458 weakly suppresses
  22. L3:1901 weakly excites
  23. L23:2003 weakly suppresses
  24. L19:816 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.

1 values between -5.79 and -5.621 values between -2.23 and -2.053 values between -1.88 and -1.701 values between -1.70 and -1.521 values between -1.34 and -1.162 values between -1.16 and -0.981 values between -0.98 and -0.812 values between -0.81 and -0.635 values between -0.63 and -0.4583 values between -0.45 and -0.27621 values between -0.27 and -0.092572 values between -0.09 and 0.08534 values between 0.08 and 0.2663 values between 0.26 and 0.4412 values between 0.44 and 0.622 values between 0.62 and 0.803 values between 0.80 and 0.981 values between 1.15 and 1.33offnegativepositiveoften

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. nhật
  2. lish
  3. enate
  4. inished
  5. oked
  6. 院
  7. ehen
  8. ULATE
  9. ikan
  10. saat

Suppresses

  1. 虢
  2. $model
  3. getService
  4. 让孩子
  5. 高尔
  6. -depend
  7. twink
  8. .Desc
  9. -needed
  10. bases