Neuron Explorer

Qwen3-0.6B

unclear: end of a phrase before a continuation

A neuron in layer 21 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 at points in the middle of a sentence where a list or a continuation is about to start, such as a comma or a word like 'from' or 'as'. When turned up, the model's text drifted toward book and paper topics, with repeated words like 'book' and 'paper'.

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

    contract has an expiration date. This is how long the option contract lives.⏎The contract also has an underlying instrument that is the subject of the contract. This can be a publicly traded stock or gold or oil or anything that is traded on the exchange

  2. #2 peak

    When it comes to protecting metal, a few options are available to create a protective, non-corrosive layer. Corrosion can ruin any steel structure, from steel girders in high

  3. #3 peak

    tea tree oil for example, is the strongest known antibiotic in the world.⏎What are Essential Oils?⏎Essential oils are concentrated oils extracted from fruit, flowers, leaves, and stems of plants. These fragrant oils help plants to attract insects

  4. #4 peak

    System⏎Students demonstrate their knowledge of connections between natural systems and the built environment in which they live.⏎Students will learn that:⏎- Humans depend on Earth's land, ocean, atmosphere, and biosphere for many different resources.⏎- Understanding natural hazards

  5. #5 peak

    also provides other information about domain names, such as mail services.⏎Why is DNS important?⏎Domain names are fundamental to the Internet. Every Internet-connected device, whether a personal computer, smart phone, or gaming console, looks up each name in the

  6. #6 peak

    chose each book and tell whether you enjoyed⏎it and what it meant to you.⏎- Read about the world around you from any two sources--books, magazines, newspapers,⏎the Internet (with your parent's permission), field

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 act of the system and the act of the system in the context of the whole system. And, at the same time, the act of the system, and the act of the system in the context of the whole system. And,

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

    1 (Bahasa Indonesia) in English-medium classrooms.⏎OVERVIEW OF BILINGUAL EDUCATION APPROACH⏎Bilingual education is one of CLIL (Content and

  2. #2 peak

    effectively. Emergency managers resolve natural disaster challenges using the Comprehensive Emergency Management (CEM) approach. CEM is broken down into four phases: mitigation, preparedness, response, and recovery (Agency, 1995).⏎Mitigation strategies

  3. #3 peak

    attention is given both to the language and the content” (Coyle et al 2010:3).⏎Historically, the idea of combining content and language learning (CLIL) comes from countries such as the USA and Australia,

  4. #4 peak

    correct after her death and is now know to<|endoftext|>Sarah Allard and Kristin Jenkins⏎Mentor: Dr. Neal Williams⏎Ecosystem services are processes and contributions from the

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. L19:877 suppresses
  2. L19:464 weakly excites
  3. L14:3025 weakly suppresses
  4. L5:2104 weakly excites
  5. L19:2569 weakly suppresses
  6. L20:932 weakly suppresses 'of' after a count or quantity
  7. L20:1651 weakly excites
  8. L13:460 weakly excites
  9. L14:496 weakly suppresses
  10. L20:2269 weakly suppresses
  11. L13:982 weakly suppresses
  12. L18:808 weakly suppresses
  13. L19:237 weakly excites
  14. L9:2218 weakly suppresses
  15. L19:1155 weakly suppresses
  16. L1:912 weakly suppresses
  17. L19:1785 weakly excites
  18. L1:2026 weakly excites
  19. L19:2198 weakly excites
  20. L14:536 weakly suppresses
  21. L12:1116 weakly suppresses 'of' in formal noun phrases, mixed
  22. L19:844 weakly excites
  23. L7:2011 weakly suppresses
  24. L18:2947 weakly excites

Feeds later layers

  1. L25:2765 excites
  2. L27:969 weakly excites
  3. L24:2391 weakly excites
  4. L26:2319 weakly excites
  5. L25:1817 weakly suppresses
  6. L22:1185 weakly excites
  7. L24:729 weakly suppresses
  8. L22:2978 weakly excites
  9. L24:2106 weakly suppresses
  10. L24:2635 weakly excites
  11. L22:778 weakly excites
  12. L24:2407 weakly suppresses
  13. L27:449 weakly excites
  14. L25:1325 weakly excites
  15. L26:752 weakly suppresses
  16. L26:1114 weakly excites
  17. L25:991 weakly excites
  18. L22:2709 weakly excites
  19. L24:1136 weakly suppresses
  20. L27:864 weakly suppresses
  21. L23:669 weakly suppresses
  22. L24:527 weakly suppresses
  23. L23:1870 weakly excites
  24. L27:1643 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.

3 values between -9.20 and -8.621 values between -8.62 and -8.051 values between -8.05 and -7.482 values between -7.48 and -6.9011 values between -6.90 and -6.3313 values between -6.33 and -5.7610 values between -5.76 and -5.1819 values between -5.18 and -4.6124 values between -4.61 and -4.0332 values between -4.03 and -3.4642 values between -3.46 and -2.8960 values between -2.89 and -2.31100 values between -2.31 and -1.74144 values between -1.74 and -1.17284 values between -1.17 and -0.591105 values between -0.59 and -0.021463 values between -0.02 and 0.55261 values between 0.55 and 1.13129 values between 1.13 and 1.7063 values between 1.70 and 2.2750 values between 2.27 and 2.8526 values between 2.85 and 3.4219 values between 3.42 and 3.998 values between 3.99 and 4.577 values between 4.57 and 5.142 values between 5.14 and 5.717 values between 5.71 and 6.295 values between 6.29 and 6.866 values between 6.86 and 7.433 values between 7.43 and 8.011 values between 8.01 and 8.582 values between 8.58 and 9.162 values between 10.30 and 10.881 values between 10.88 and 11.451 values between 12.60 and 13.171 values between 13.17 and 13.74offnegativepositiveoften

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. curtains
  2. -books
  3. 电线
  4. 书籍
  5. wives
  6. papers
  7. Paper
  8. книги
  9. 图书
  10. books

Suppresses

  1. immune
  2. anst
  3. ixa
  4. resource
  5. 风险管理
  6. TES
  7. activity
  8. ENTIAL
  9. проц
  10. 活动