Neuron Explorer

Qwen3-0.6B

opening bracket before an acronym

A neuron in layer 24 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 lights up on the opening bracket or first letters just before a short abbreviation is spelled out, such as the "(" in "deep belief networks (DBN)" or "Rural Fire Service (RFS)". When it was turned up, the model's writing became unrelated and rambling, so the steering shows no clear effect.

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

    So, for a language to be type safe, we have to provide another alternative to the CTS, which is the Common Language Specification.⏎- Common Language Specifications (CLS):⏎CLS is a set of rules that describes, in detail, the minimal

  2. #2 peak

    colour of the mineral using crossed polars).⏎Minerals have different characteristics using the different types of light – plain polarised (PPL) and cross polarised (XPL) . As an example, in a Cornubian granite, the

  3. #3 peak

    ratified later in 1865, began<|endoftext|>In late December, at the very peak of our region’s fire emergency, somebody rang our Rural Fire Service (RFS) headquarters and complained: ‘Why haven’t you put the fires out

  4. #4 peak

    innovative programming such as the Family Innovative Supportive Housing (FISH) Program. A collaborative effort between the Children's Home of York, Y Community Development Corporation (YCDC), and the York County Office of Children, Youth, and Families (OC

  5. #5 peak

    .⏎As reported by Overclock3D; the study results are based on the widely accepted Greenhouse Gas Protocol [GHGP] established by the World Resources Institute [WRI] and World Business Council for Sustainable Development [WBCSD and were

  6. #6 peak

    fuel cell car fleet may hit top gear within five years, but only if there is enough investment in filling stations, the UK Hydrogen and Fuel Cells Association (UK HFCA) told Reuters on Friday.⏎Fuel cells convert hydrogen into electricity, with

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: how to use the properties of real or virtual characters (e.g., character loop loops or character loops) vertically horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally horizontally

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

    MS, SD, WI, WV, and, VT). Roughly a third of Hispanic women were poor in six states (IN, MS, NC, PA, RI, TN).⏎Yet again, these numbers underscore the importance of federal and state

  2. #2 peak

    University of Otago’s Department of Psychology, they performed this feat by tracking the statistical likelihood that “bigrams,” letter pairs such as “EN” and “AL,” were more likely associated with words or non-

  3. #3 peak

    Live kids. Need I say more?<|endoftext|>New life breathed new life into Woodford County’s historic African-American community⏎WOODFORD COUNTY, Ky. (WKYT) – Ten years after the last resident left a historic African American community in

  4. #4 peak

    ;⏎P0 = 0xB0;⏎P0 = 0x99;⏎P0 = 0x92;⏎P0 = 0x82;⏎P0 = 0xF8;⏎P0 = 0

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. L21:496 strongly excites
  2. L22:1393 suppresses
  3. L23:404 suppresses
  4. L19:1778 excites
  5. L21:732 suppresses
  6. L18:2520 excites
  7. L23:326 excites
  8. L23:672 excites
  9. L20:2327 suppresses
  10. L22:839 excites
  11. L21:1594 excites
  12. L21:306 excites
  13. L23:383 suppresses
  14. L20:1338 excites
  15. L19:1781 weakly suppresses
  16. L20:1675 weakly suppresses
  17. L17:2755 weakly suppresses
  18. L20:2727 weakly suppresses
  19. L20:680 weakly suppresses
  20. L21:1355 weakly suppresses
  21. L19:2045 weakly excites abbreviation pieces in organisation names
  22. L19:1011 weakly excites
  23. L22:1705 weakly excites
  24. L19:761 weakly suppresses

Feeds later layers

  1. L27:340 excites parenthesis introducing an acronym
  2. L26:257 excites
  3. L27:830 excites
  4. L26:1756 suppresses
  5. L26:45 suppresses
  6. L26:1235 excites opening parenthesis before an acronym
  7. L25:1105 suppresses
  8. L27:1072 excites
  9. L26:380 weakly excites
  10. L27:229 weakly suppresses
  11. L27:1009 weakly excites opening parenthesis before an acronym
  12. L26:293 weakly excites
  13. L25:522 weakly excites
  14. L27:1781 weakly excites
  15. L27:1059 weakly excites
  16. L27:1210 weakly suppresses
  17. L25:354 weakly suppresses
  18. L26:219 weakly suppresses
  19. L25:687 weakly excites
  20. L25:2623 weakly suppresses
  21. L25:673 weakly suppresses
  22. L27:412 weakly suppresses
  23. L26:2162 weakly suppresses
  24. L27:1453 weakly suppresses

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 -9.04 and -8.421 values between -7.19 and -6.572 values between -5.33 and -4.711 values between -4.71 and -4.101 values between -4.10 and -3.482 values between -3.48 and -2.865 values between -2.86 and -2.245 values between -2.24 and -1.6211 values between -1.62 and -1.0196 values between -1.01 and -0.393086 values between -0.39 and 0.23633 values between 0.23 and 0.8552 values between 0.85 and 1.472 values between 1.47 and 2.082 values between 2.08 and 2.701 values between 2.70 and 3.322 values between 3.94 and 4.561 values between 5.79 and 6.411 values between 7.03 and 7.651 values between 8.26 and 8.881 values between 12.59 and 13.211 values between 15.06 and 15.68offnegativepositiveoften

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. lando
  2. eur
  3. abc
  4. salsa
  5. ddd
  6. ccc
  7. XYZ
  8. efe
  9. ronics
  10. ABC

Suppresses

  1. horizontally
  2. YM
  3. YL
  4. YT
  5. ys
  6. orrow
  7. YT
  8. DS
  9. getObject
  10. antage