Neuron Explorer

Qwen3-0.6B

citation and catalogue metadata text

A neuron in layer 12 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the reference-list and catalogue parts of web pages, such as journal names, page ranges, DOIs, and publication status lines. It is most negative on commas inside numbers and lists. Turning it up made the written text wander into random topics, so the steering results do not show a clear theme.

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

    andfonline.com/pricing/journal/VECE20|⏎When requesting a correction, please mention this item's handle: RePEc:taf:jeduce:v:42:y:2011:i:2:p

  2. #2 peak

    lives. If women united internationally and with the wider struggles of working<|endoftext|>|Part of a series on|⏎|Types of creationism|⏎|Other religious views|⏎The Omph

  3. #3 peak

    23(2), 1976 pp. 125–135. doi:10.1112/S002557930000871

  4. #4 peak

    ological event horizon. We anticipate that this apparent violation of the generalized second law will disappear when solutions are available for black holes embedded in arbitrary backgrounds.⏎|Original language||English (US)|⏎|Number of pages||12|⏎|Journal||

  5. #5 peak

    students learning in their mother tongue are some factors for consideration. Copyright © 2014 Society for the Scientific Study of Reading (SSSR).⏎|Publication status||Published - Jul

  6. #6 peak

    1: Download Drupal⏎First download the Drupal using following command:⏎# wget http://ftp.drupal.org/files/projects/drupal-7.33.zip⏎Step 2: Extract

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 the one who will be the one who will be the one who will be the one who will be the one who will be the one who will be the one who will be the one who will be the one who will be 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

    -oh-sen-TEE-sis) or chorionic (pronounced kawr-ee-ON-ik) villus sampling.1,2,3⏎- Amniocentesis. This involves inserting a needle into the

  2. #2 peak

    ATIC COMPOUNDS ARE THOSE ORGANIC COMPOUNDS WHICH HAVE BENZENE RING (CYCLOHEXA-1,3,5-TRIENE). AROMATIC COMPOUNDS ARE ALWAYS F

  3. #3 peak

    ingredient thallium acetate.(3) Thallium acetate is used to kill rats and was known to cause baldness, pain, and paralysis.(3) A study done at the University of California revealed that people working in the cosmetics

  4. #4 peak

    due to its antibacterial and anti-inflammatory properties. Other medicinal uses include using wild chervil to treat cancer, kidney stones, and cystitis.1,2⏎Cultivation and harvest:⏎Wild chervil has become naturalized across

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. L7:809 weakly suppresses
  2. L8:18 weakly suppresses
  3. L9:2489 weakly suppresses
  4. L0:2246 weakly excites
  5. L3:110 weakly suppresses
  6. L6:1700 weakly excites
  7. L5:1400 weakly excites
  8. L4:97 weakly suppresses
  9. L5:1498 weakly excites
  10. L10:1241 weakly suppresses
  11. L4:355 weakly suppresses
  12. L3:1861 weakly suppresses
  13. L7:738 weakly excites
  14. L9:5 weakly excites
  15. L8:10 weakly excites
  16. L6:1711 weakly excites
  17. L1:2212 weakly suppresses
  18. L0:170 weakly suppresses
  19. L6:322 weakly excites
  20. L6:2771 weakly excites
  21. L7:2702 weakly suppresses
  22. L7:1913 weakly suppresses
  23. L1:1551 weakly excites
  24. L10:247 weakly suppresses

Feeds later layers

  1. L14:1984 suppresses
  2. L13:758 suppresses 'and' joining a pair of options
  3. L15:691 weakly suppresses
  4. L15:677 weakly suppresses
  5. L13:2294 weakly excites
  6. L19:1823 weakly suppresses 'and' inside organisation or paired names
  7. L17:1669 weakly suppresses
  8. L15:2722 weakly suppresses "and" or "or" in a pair or list
  9. L15:2524 weakly excites
  10. L19:1414 weakly suppresses
  11. L26:1278 weakly excites
  12. L26:1853 weakly excites
  13. L25:1592 weakly excites
  14. L15:2406 weakly suppresses
  15. L16:943 weakly excites
  16. L16:511 weakly suppresses
  17. L18:1429 weakly excites
  18. L14:1804 weakly suppresses "and" in subject-area lists
  19. L13:2330 weakly suppresses
  20. L23:2978 weakly excites
  21. L13:462 weakly excites
  22. L13:826 weakly excites
  23. L14:2463 weakly suppresses
  24. L21:622 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 -3.58 and -3.421 values between -3.26 and -3.102 values between -1.33 and -1.173 values between -1.17 and -1.011 values between -1.01 and -0.853 values between -0.85 and -0.693 values between -0.69 and -0.5238 values between -0.52 and -0.36246 values between -0.36 and -0.20918 values between -0.20 and -0.041610 values between -0.04 and 0.12804 values between 0.12 and 0.28234 values between 0.28 and 0.4437 values between 0.44 and 0.603 values between 0.60 and 0.761 values between 0.76 and 0.931 values between 1.09 and 1.251 values between 1.41 and 1.571 values between 2.70 and 2.86offnegativepositiveoften

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. gar
  2. enburg
  3. cala
  4. 似
  5. obe
  6. goals
  7. 격
  8. caster
  9. 冢
  10. casting

Suppresses

  1. ijke
  2. dek
  3. ,opt
  4. 高效的
  5. 펌
  6. 这两种
  7. ählt
  8. ANNOT
  9. McGregor
  10. ennis