Neuron Explorer

Qwen3-0.6B

hyphen joining two words

A neuron in layer 17 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 fires on the hyphen that joins two words into one compound, such as "cold-weather" or "tobacco-cessation." Its weaker opposite side fires on minus signs in numbers. Turning it up or down gave text that was not clearly about hyphens.

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

    know it (more or less). Thus began the era of dancing babies and other horrors best forgotten<|endoftext|>PROTEOMICS is the study of many proteins at once- or a description of proteins in a cell, tissue or cellular organelle on

  2. #2 peak

    . Hernias can be painful and limit the mobility of pigs.⏎- Having<|endoftext|>This time of year, it may be difficult to detect the cause of your cold-weather sniffles. Colds, sinus infections and the flu all share many symptoms

  3. #3 peak

    MIT News Office. 1992-04-15. http://web.mit.edu/newsoffice/1992/citation-0415.html. Retrieved 200<|endoftext|>Our herd of awesome

  4. #4 peak

    the participants.⏎People generally didn’t mind being touched by Cody overall, but were less comfortable with the robot when it spoke to them beforehand. And participants were more-accepting of a potentially necessary medical touch than an attempt at a soothing touch by

  5. #5 peak

    48 percent in Hispanics); and less likely to have used tobacco-cessation aids during the past year in a quit attempt (38 percent in

  6. #6 peak

    that, it is the way women-issues are treated at the working place that translates into them earning less than their male

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: - - _ 1. 1000 1000 1000

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

    i Marincas | Shutterstock⏎Atomic Number: 86⏎Atomic Symbol: Rn⏎Atomic Weight: 222⏎Melting Point: -95 F (-71 C)⏎Boiling Point: -79 F

  2. #2 peak

    This example demonstrates different noise distributions methods that can be selected for a Noise block. Both noise blocks use samplePeriod = 0.02 s, y_min=-1, y_max=3, and have identical fixedLocalSeed. This means

  3. #3 peak

    In July, the warmest month, the average summit temperature is -19° C (-2° F).⏎When George Mallory, the British

  4. #4 peak

    40°F (-30°C/40°C) Other cover qualities (such as oil or fire) are usually only able to withstand a minimum temperature of -20°C. For temperatures lower than this, conveyors should be fitted with

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. L4:46 strongly suppresses
  2. L16:844 strongly suppresses
  3. L0:1703 strongly excites
  4. L3:101 suppresses
  5. L0:848 excites
  6. L16:893 suppresses
  7. L1:320 excites
  8. L15:241 excites
  9. L16:264 excites opening parenthesis in citations and asides
  10. L7:89 excites dash in year or number ranges
  11. L6:338 suppresses
  12. L1:38 suppresses
  13. L7:432 excites
  14. L5:1304 excites
  15. L1:1336 excites
  16. L0:1267 suppresses
  17. L10:823 suppresses
  18. L0:1128 suppresses
  19. L5:291 weakly excites hyphens inside codes and page ranges
  20. L10:156 weakly suppresses
  21. L14:415 weakly suppresses
  22. L2:317 weakly suppresses
  23. L10:134 weakly excites
  24. L7:122 weakly excites

Feeds later layers

  1. L18:691 strongly excites
  2. L19:804 strongly excites
  3. L20:1886 strongly excites
  4. L19:1879 suppresses
  5. L19:2267 excites hyphen joining compound words
  6. L18:1142 suppresses
  7. L18:642 suppresses
  8. L26:2003 excites hyphen joining two word parts
  9. L23:1479 excites
  10. L18:1541 suppresses
  11. L22:2211 excites
  12. L23:793 weakly suppresses
  13. L18:2369 weakly suppresses
  14. L19:1398 weakly suppresses
  15. L18:1047 weakly excites
  16. L19:171 weakly suppresses
  17. L24:266 weakly suppresses
  18. L25:2347 weakly excites
  19. L18:713 weakly suppresses
  20. L27:504 weakly suppresses
  21. L20:183 weakly suppresses
  22. L25:821 weakly excites
  23. L20:1571 weakly excites
  24. L26:811 weakly excites hyphen in number or size compounds

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.

130 values between -1.50 and -0.892125 values between -0.89 and -0.281573 values between -0.28 and 0.3416 values between 0.34 and 0.956 values between 0.95 and 1.565 values between 1.56 and 2.174 values between 2.17 and 2.785 values between 2.78 and 3.397 values between 3.39 and 4.004 values between 4.00 and 4.622 values between 4.62 and 5.236 values between 5.23 and 5.842 values between 6.45 and 7.061 values between 7.67 and 8.282 values between 8.28 and 8.892 values between 8.89 and 9.512 values between 10.12 and 10.732 values between 11.34 and 11.951 values between 11.95 and 12.561 values between 12.56 and 13.171 values between 13.17 and 13.781 values between 13.78 and 14.402 values between 15.01 and 15.621 values between 16.23 and 16.841 values between 16.84 and 17.451 values between 17.45 and 18.061 values between 18.06 and 18.672 values between 18.67 and 19.291 values between 21.73 and 22.341 values between 22.34 and 22.95offnegativepositiveoften

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. -Identifier
  2. ual
  3. NOWLED
  4. çe
  5. ysis
  6. ance
  7. rench
  8. èle
  9. nowled
  10. addr

Suppresses

  1. раг
  2. عيد
  3. шел
  4. Form
  5. Forms
  6. den
  7. +"]
  8. bur
  9. 发
  10. 齊