Neuron Explorer

Qwen3-0.6B

unclear: line breaks and list dashes

A neuron in layer 10 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 reacts to line breaks and list markers in messy web text, like the start of a new line in a bibliographic record or a dash starting a bullet. The examples don't share a clear meaning. Turning it up gave loose, repetitive text that didn't clearly relate to the neuron's triggers.

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

    conservation area.⏎Neil burn Drive was created.⏎Canmore place was erected.<|endoftext|>ERIC Number: ED379961⏎Record Type: RIE⏎Publication Date: 19

  2. #2 peak

    and find out how you can connect with other literacy professionals.⏎Teacher Resources by Grade⏎|1st - 2nd||3rd - 4th|⏎|5th - 6th||7th - 8th|⏎|9

  3. #3 peak

    :⏎- establish a Hunger-Free Task Force;⏎- designate staff to connect eligible students with the

  4. #4 peak

    :⏎- Rabbits are herbivores which mean they only eat plants.⏎- Bunn

  5. #5 peak

    :⏎- Vitamin E (α-tocopherol)⏎- Causes and Risks for Vitamin

  6. #6 peak

    :⏎- Interpersonal, project management, risk management and time management skills.⏎- Software development

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 Great Gatsby," I can't even begin to imagine the scope of what's going on in this nation's political turmoil, and the National Socialist, Germany, and defeated state of Washington in a country that lost the war to the

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

    string, as well as the number of characters in the string variable⏎>>> len("a string") 8 >>> >>> s = "a long string" >>> >>> len(s) 13 >>> >>> len("") 0 >>>⏎>>> >>>

  2. #2 peak

    5 volume without having to “zero out” the data blocks. To do so, do one of the following:⏎Use the metainit command with the -k option. The -k option recreates the RAID 5 volume without initializing

  3. #3 peak

    Standard▼▲⏎- Grade Level▼▲⏎- Media Type▼▲⏎- Guides & Workbooks▼▲⏎- Author / Artist▼▲⏎- Top Rated▼▲⏎Guide

  4. #4 peak

    to close the dialog|⏎|·||Is used in more than one diagram, the Element Usage dialog displays listing

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. L9:542 weakly excites
  2. L4:194 weakly suppresses
  3. L9:936 weakly suppresses
  4. L5:47 weakly excites
  5. L8:16 weakly suppresses
  6. L8:14 weakly excites
  7. L7:2971 weakly suppresses
  8. L4:183 weakly excites
  9. L4:1950 weakly excites
  10. L8:10 weakly excites
  11. L5:238 weakly excites
  12. L7:841 weakly suppresses
  13. L2:1351 weakly excites
  14. L5:1289 weakly excites
  15. L4:729 weakly excites
  16. L1:1122 weakly excites
  17. L2:1319 weakly excites
  18. L9:2061 weakly suppresses
  19. L1:2687 weakly suppresses
  20. L5:1270 weakly suppresses
  21. L5:1993 weakly excites
  22. L0:907 weakly excites
  23. L7:1650 weakly excites
  24. L7:303 weakly suppresses

Feeds later layers

  1. L12:1144 excites
  2. L11:2434 weakly excites
  3. L13:2771 weakly excites
  4. L14:2019 weakly suppresses
  5. L11:1093 weakly excites
  6. L22:1463 weakly excites
  7. L13:1572 weakly suppresses
  8. L22:2621 weakly excites
  9. L23:1295 weakly excites
  10. L17:1949 weakly suppresses
  11. L19:1537 weakly excites
  12. L25:2589 weakly suppresses
  13. L18:2213 weakly excites
  14. L16:1809 weakly suppresses comma inside a list of names or acronyms
  15. L20:288 weakly suppresses
  16. L24:11 weakly suppresses
  17. L19:1422 weakly excites
  18. L11:2947 weakly excites
  19. L12:678 weakly excites
  20. L14:2521 weakly excites
  21. L20:615 weakly excites
  22. L15:2270 weakly excites
  23. L12:555 weakly suppresses
  24. L24:1796 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 -0.57 and -0.541 values between -0.54 and -0.513 values between -0.48 and -0.451 values between -0.45 and -0.428 values between -0.42 and -0.397 values between -0.39 and -0.3610 values between -0.36 and -0.3324 values between -0.33 and -0.3025 values between -0.30 and -0.2737 values between -0.27 and -0.2447 values between -0.24 and -0.2175 values between -0.21 and -0.18101 values between -0.18 and -0.15105 values between -0.15 and -0.12193 values between -0.12 and -0.09212 values between -0.09 and -0.06304 values between -0.06 and -0.03351 values between -0.03 and -0.00406 values between -0.00 and 0.03393 values between 0.03 and 0.06333 values between 0.06 and 0.09293 values between 0.09 and 0.11227 values between 0.11 and 0.14221 values between 0.14 and 0.17166 values between 0.17 and 0.20117 values between 0.20 and 0.2375 values between 0.23 and 0.2653 values between 0.26 and 0.2941 values between 0.29 and 0.3218 values between 0.32 and 0.3518 values between 0.35 and 0.3814 values between 0.38 and 0.418 values between 0.41 and 0.444 values between 0.44 and 0.476 values between 0.47 and 0.502 values between 0.50 and 0.532 values between 0.53 and 0.564 values between 0.56 and 0.592 values between 0.59 and 0.62offnegativepositiveoften

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. very
  2. tó
  3. ыта
  4. ⏎⏎⏎
  5. remely
  6. ⏎⏎⏎⏎
  7. зависимости
  8. струк
  9. ragment
  10. atives

Suppresses

  1. oproject
  2. RIA
  3. cuenta
  4. erta
  5. cratch
  6. 打入
  7. isme
  8. gå
  9. ollapsed
  10. 兴旺