Neuron Explorer

Qwen3-0.6B

text boundary after a document break

A neuron in layer 9 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 the invisible marker that separates two unrelated web pages, especially right where a new page starts. Turning it up didn't produce a clear change in the writing.

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

    made many<|endoftext|>NSCAW, No. 19: Risk of Long-term Foster Care

  2. #2 peak

    959,<|endoftext|>Everyone's heard about an apple a day. As one of the most popular fruits

  3. #3 peak

    ,0<|endoftext|>Guidelines for Restoring Cultural Landscapes⏎Identifying, retaining and preserving

  4. #4 peak

    7Availability: In StockCBD Stock No: WW265132<|endoftext|>Ophthalmologists work in hospitals and clinics on land and aboard ships.⏎Average Military

  5. #5 peak

    718<|endoftext|>Women who have kids later on in life may live longer, according to the findings

  6. #6 peak

    0240.<|endoftext|>By 2003, plenty of data suggested that pesticide drift was harming

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 want to add to the user's request. The user's request is to build a new application, and create a new website. The user has asked to create a new website, and I want to build a new application. So,

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

    soon after), it’s time the real experts divulge the dirty, little secret of Lincoln studies: We keep finding new evidence about the life of the 16th president, and some of it can be kind of shocking.⏎It’s worth

  2. #2 peak

    less need for water, fertilizer and pesticides. Also, native plants are less likely to create environmental havoc should they "escape" from your garden, unlike aggressive invasives.⏎Native plants have an important role in the local ecosystem (community

  3. #3 peak

    along the road to recovery. The rebound follows cuts to global emissions of chlorofluorocarbons and halons, which have been destroying the gase

  4. #4 peak

    scientific and industrial communities have collaborated to find safe and economical replacements for chlorofluorocarbons (CFCs), the manmade chemicals implicated in

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. L5:2 strongly excites
  2. L8:2395 strongly suppresses
  3. L8:257 strongly excites
  4. L8:67 strongly excites
  5. L8:50 suppresses
  6. L8:291 suppresses
  7. L3:5 suppresses
  8. L8:341 excites
  9. L5:135 excites
  10. L3:63 suppresses
  11. L7:3 excites
  12. L3:551 excites
  13. L8:2765 suppresses
  14. L7:1902 excites
  15. L8:136 excites
  16. L0:9 weakly excites
  17. L3:604 weakly suppresses
  18. L7:2770 weakly suppresses
  19. L3:133 weakly excites
  20. L4:382 weakly excites
  21. L3:139 weakly excites
  22. L6:1800 weakly suppresses
  23. L1:853 weakly excites
  24. L2:322 weakly suppresses

Feeds later layers

  1. L20:96 strongly excites
  2. L20:116 strongly suppresses
  3. L27:118 strongly suppresses
  4. L22:156 strongly suppresses
  5. L19:552 strongly suppresses
  6. L10:32 strongly suppresses
  7. L20:429 strongly suppresses
  8. L26:120 strongly suppresses
  9. L26:222 strongly suppresses
  10. L26:312 strongly suppresses
  11. L19:721 strongly suppresses
  12. L21:270 strongly suppresses
  13. L22:115 strongly excites
  14. L21:269 strongly excites
  15. L26:266 suppresses
  16. L18:109 suppresses
  17. L25:323 suppresses
  18. L25:251 suppresses
  19. L26:595 suppresses
  20. L26:412 suppresses
  21. L27:104 suppresses
  22. L26:29 suppresses
  23. L10:70 suppresses
  24. L16:270 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 -3.90 and -3.711 values between -3.52 and -3.341 values between -3.34 and -3.152 values between -3.15 and -2.961 values between -2.96 and -2.787 values between -2.78 and -2.598 values between -2.59 and -2.4015 values between -2.40 and -2.2227 values between -2.22 and -2.0332 values between -2.03 and -1.8441 values between -1.84 and -1.6683 values between -1.66 and -1.4799 values between -1.47 and -1.28139 values between -1.28 and -1.10217 values between -1.10 and -0.91251 values between -0.91 and -0.72342 values between -0.72 and -0.54407 values between -0.54 and -0.35412 values between -0.35 and -0.16469 values between -0.16 and 0.03410 values between 0.03 and 0.21316 values between 0.21 and 0.40207 values between 0.40 and 0.58137 values between 0.58 and 0.77110 values between 0.77 and 0.9660 values between 0.96 and 1.1441 values between 1.14 and 1.3327 values between 1.33 and 1.5218 values between 1.52 and 1.708 values between 1.70 and 1.894 values between 1.89 and 2.088 values between 2.08 and 2.262 values between 2.26 and 2.452 values between 2.45 and 2.642 values between 2.82 and 3.011 values between 3.38 and 3.57offnegativepositiveoften

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. .instagram
  2. orally
  3. Catalonia
  4. agreeing
  5. Carn
  6. Maced
  7. BED
  8. arresting
  9. contacted
  10. Hampton

Suppresses

  1. 字段
  2. ユーザー
  3. User
  4. backend
  5. user
  6. Object
  7. cache
  8. ⏎
  9. データ
  10. 程序员