Neuron Explorer

Qwen3-0.6B

comma inside numbers

A neuron in layer 3 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 responds to the comma in big numbers, like the one in 3,789 or 70,000. Turning it up did not produce any clear change in the written text.

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

    1.800 square kilometres big. It includes mountain ranges along the central Alps with Austria′s highest mountain, the Großglockner (3,789 metres), some 130 square kilometres of glaciers (even

  2. #2 peak

    BTW € 125,00<|endoftext|>Hell Maximillian (1720-1792

  3. #3 peak

    |Titus Flavius Vespasianus||Simon Bar Giora|| John of Gischala|⏎Eleazar ben Simon⏎|70,000 men||20-30,000 men||

  4. #4 peak

    risk factors. Mark Goldstein, 79,<|endoftext|>Sea surface temperature⏎Sea surface temperature (SST) is the water temperature

  5. #5 peak

    platform MijnTuinlab (My Garden lab).⏎The project was enthusiastically received, as by the end of September the perimeters of more than 10,000 gardens had already been registered and for more than 1,1

  6. #6 peak

    con células madre embrionarias humanas revela los primeros pasos en el desarrollo humanoContributed by: AdminStaff1 · Views: 1,130⏎Contributed by: AdminStaff1 · April 10

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 have been thinking about lately about the fact that I have been thinking about lately about the fact that I have been thinking about lately. I have been thinking about lately about the fact that I have been thinking about lately about the fact that I

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

    book. What about the problems we face in adulthood? (read more here)<|endoftext|>Задание от гостя:Размещено 13.09.2016 19:19

  2. #2 peak

    -Institut für klinische Hirnforschung (HIH)⏎CRISPR meets single-cell sequencing in new screening method⏎19.01.2017 | CeMM Forschungszentrum für

  3. #3 peak

    re de Kerguelen, by Rigobert Bonne, from Atlas Encyclopedique, circa 1787, (11.3 x 9.5"). This copper engraved map depicts the Kerguel

  4. #4 peak

    1 : 10101100.00010000.0yyyyyyy.yyyyyyyy => 172.16.0.0/17 SUBNET2 : 10

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. L0:1224 suppresses
  2. L0:1301 suppresses
  3. L0:178 suppresses
  4. L2:166 suppresses
  5. L1:2055 weakly suppresses
  6. L2:36 weakly excites
  7. L2:1 weakly suppresses
  8. L2:570 weakly suppresses
  9. L2:1710 weakly suppresses
  10. L1:2677 weakly suppresses
  11. L2:111 weakly suppresses
  12. L1:1719 weakly suppresses
  13. L1:2412 weakly excites
  14. L1:2342 weakly excites
  15. L0:306 weakly suppresses
  16. L1:274 weakly suppresses
  17. L0:248 weakly suppresses
  18. L2:75 weakly suppresses
  19. L0:215 weakly suppresses
  20. L1:830 weakly suppresses
  21. L0:2466 weakly suppresses
  22. L1:1065 weakly suppresses
  23. L0:1193 weakly excites
  24. L2:1251 weakly suppresses

Feeds later layers

  1. L5:582 excites
  2. L4:440 excites
  3. L4:1867 suppresses
  4. L8:2406 weakly suppresses
  5. L15:1446 weakly suppresses
  6. L18:2607 weakly suppresses
  7. L16:1083 weakly excites
  8. L23:585 weakly excites
  9. L19:2906 weakly suppresses
  10. L24:1070 weakly excites
  11. L4:669 weakly suppresses
  12. L7:2537 weakly suppresses
  13. L16:2849 weakly excites
  14. L13:421 weakly suppresses
  15. L4:458 weakly suppresses
  16. L6:1700 weakly suppresses
  17. L13:2051 weakly suppresses
  18. L22:858 weakly excites
  19. L22:2894 weakly suppresses
  20. L8:2745 weakly suppresses
  21. L23:2053 weakly excites
  22. L5:2001 weakly suppresses
  23. L17:2615 weakly suppresses
  24. L10:2913 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 -0.49 and -0.471 values between -0.47 and -0.452 values between -0.45 and -0.432 values between -0.43 and -0.403 values between -0.38 and -0.364 values between -0.36 and -0.3310 values between -0.33 and -0.318 values between -0.31 and -0.2912 values between -0.29 and -0.2615 values between -0.26 and -0.2416 values between -0.24 and -0.2229 values between -0.22 and -0.2047 values between -0.20 and -0.1743 values between -0.17 and -0.1585 values between -0.15 and -0.13137 values between -0.13 and -0.10174 values between -0.10 and -0.08249 values between -0.08 and -0.06328 values between -0.06 and -0.03514 values between -0.03 and -0.01833 values between -0.01 and 0.01561 values between 0.01 and 0.04319 values between 0.04 and 0.06194 values between 0.06 and 0.08125 values between 0.08 and 0.1070 values between 0.10 and 0.1343 values between 0.13 and 0.1519 values between 0.15 and 0.1717 values between 0.17 and 0.2015 values between 0.20 and 0.2212 values between 0.22 and 0.246 values between 0.24 and 0.276 values between 0.27 and 0.294 values between 0.29 and 0.311 values between 0.31 and 0.332 values between 0.36 and 0.381 values between 0.40 and 0.43offnegativepositiveoften

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. $link
  2. .sock
  3. Design
  4. manship
  5. culus
  6. proposals
  7. Outer
  8. outward
  9. THON
  10. AppComponent

Suppresses

  1. ões
  2. uten
  3. egas
  4. ão
  5. ugin
  6. GEST
  7. orz
  8. ZE
  9. ena
  10. uada