Neuron Explorer

Qwen3-0.6B

'while' joining two clauses after a comma

A neuron in layer 6 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 on the word 'while' when it comes after a comma and links two contrasting parts of a sentence. Turning it up did not give a clearly related change in the replies.

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

    ⏎There are many types of batteries that you might use on your camper or RV, each with its advantages and disadvantages. Some are better suited for winter camping, while some are better for boondocking. It's important to know the type of

  2. #2 peak

    ticket, which can be done online or at any store that sells state-approved tickets. Ticket prices vary, but most are inexpensive. Some are scratch-off cards, while others are more expensive and require a computer to play. Some people also buy a

  3. #3 peak

    . Some foods can trigger inflammation or allergic reactions that can exacerbate the symptoms of it. Spicy foods, for example, can cause redness and flushing, while acidic foods can cause irritation and dryness. Dairy products and processed

  4. #4 peak

    During testing, control nets caught 125 green turtles, while the LED-lit nets ensnared just 62, with no negative

  5. #5 peak

    difference between the weather in the UK and southern Europe. In the UK we have on average about 1000+ hours of sunshine per year, where as in southern Europe it's more like 2,500-30

  6. #6 peak

    States, while the Navajo still suffer as one of the most impoverished communities in the country.⏎

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 I can be as effective as the first one, and I can be as effective as the first one, and I can be as effective as the first one, and I can be as effective as the first one. Wait, but I

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

    that the camera could be plugged in directly in the wall outlet in the different measurement rooms.⏎The light weight and small size of the array ensured easy portability and also great flexibility in measurement positions. The camera could be set up on the tripod,

  2. #2 peak

    tests or grades or awards: curiosity is always a finer spur than rivalry. Nor do we diminish the students’ discoveries by reciting the famous names of those who

  3. #3 peak

    functions in the body for maintaining life. Diseases such as cancer are not only characterized by altered genes, but also by disturbed protein production. But how is protein production controlled?⏎Researchers of the

  4. #4 peak

    the amount of sodium balance. Orange keeps blood pressure correct by keeping sodium content normal. That is why anyone who has trouble with high or low

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. L4:486 excites
  2. L0:1042 weakly suppresses "This" starting a labelled sentence
  3. L4:2835 weakly excites semicolons joining two clauses
  4. L1:1401 weakly suppresses
  5. L5:2359 weakly excites
  6. L2:1165 weakly excites
  7. L1:725 weakly suppresses "to" in "referred to"
  8. L0:565 weakly excites
  9. L2:305 weakly excites
  10. L3:1607 weakly suppresses
  11. L3:2140 weakly suppresses
  12. L0:2020 weakly excites
  13. L0:993 weakly suppresses
  14. L5:1124 weakly excites
  15. L2:1696 weakly suppresses
  16. L0:1075 weakly excites
  17. L5:2426 weakly suppresses
  18. L3:2123 weakly suppresses
  19. L2:144 weakly excites
  20. L3:1423 weakly suppresses
  21. L0:2533 weakly suppresses
  22. L3:1491 weakly suppresses
  23. L2:353 weakly excites
  24. L2:2656 weakly suppresses

Feeds later layers

  1. L9:1374 excites
  2. L15:2671 excites 'and' joining paired items
  3. L9:1089 excites
  4. L11:1242 excites
  5. L15:2310 excites
  6. L13:933 excites
  7. L18:2862 excites
  8. L17:2427 excites
  9. L12:972 excites
  10. L15:545 excites
  11. L9:1170 excites
  12. L13:125 excites
  13. L14:2150 excites
  14. L17:1499 weakly excites
  15. L12:1120 weakly excites
  16. L17:806 weakly suppresses
  17. L11:682 weakly excites gap before a second number
  18. L10:1898 weakly excites
  19. L18:2838 weakly excites
  20. L12:1378 weakly excites
  21. L8:70 weakly excites 'and' between stacked superlatives
  22. L12:141 weakly suppresses
  23. L22:1988 weakly suppresses "the" in "the more..., the more" phrases
  24. L18:1955 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 -1.79 and -1.711 values between -1.71 and -1.631 values between -1.38 and -1.301 values between -1.30 and -1.221 values between -1.22 and -1.141 values between -1.14 and -1.052 values between -1.05 and -0.972 values between -0.89 and -0.811 values between -0.81 and -0.731 values between -0.73 and -0.647 values between -0.64 and -0.567 values between -0.56 and -0.4819 values between -0.48 and -0.4047 values between -0.40 and -0.31101 values between -0.31 and -0.23244 values between -0.23 and -0.15516 values between -0.15 and -0.071106 values between -0.07 and 0.02911 values between 0.02 and 0.10491 values between 0.10 and 0.18254 values between 0.18 and 0.2691 values between 0.26 and 0.3441 values between 0.34 and 0.4320 values between 0.43 and 0.518 values between 0.51 and 0.597 values between 0.59 and 0.677 values between 0.67 and 0.765 values between 0.76 and 0.842 values between 0.84 and 0.922 values between 0.92 and 1.004 values between 1.00 and 1.083 values between 1.08 and 1.172 values between 1.17 and 1.251 values between 1.41 and 1.49offnegativepositiveoften

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. rote
  2. itize
  3. rif
  4. owers
  5. -vertical
  6. ROID
  7. rowsers
  8. 另有
  9. MENT
  10. Vul

Suppresses

  1. 径
  2. ))->
  3. )object
  4. )).
  5. ęż
  6. ា
  7. .)
  8. ))*
  9. ))==
  10. vetica