Neuron Explorer

Qwen3-0.6B

blank space before a number

A neuron in layer 26 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 empty gap just before a number, such as the space in "on July 19-20" or "from 1332 BC", where the number has not started yet. Turned down, the model produced only blank lines, so the neuron seems tied to whitespace.

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

    urons are ‘brains of the brain’, so to speak, and the rest of the cells were thought to be there only for support. But neurons account for only 15 percent of the brain, while these so-called ‘support’ cells occupy

  2. #2 peak

    and medical personnel; and U.S. Navy and Coast Guard ships are deployed to support launch contingencies and astronaut recovery. Additionally, the Navy provides a KC-130 tanker for helicopter air refueling, E-2C aircraft for enhanced

  3. #3 peak

    ges 8-11)⏎The Seneca Falls Convention was a meeting of supporters of women’s rights in the United States. The meeting took place on July 19-20, 1848, at Seneca Falls

  4. #4 peak

    , death will eventually take us all.⏎However, some people don't look like they're dead at all. Read on!⏎Tutankhamun lived from 1332 BC to 1323 BC. He is widely

  5. #5 peak

    politics” to the impact of the “digital revolution,” the course looks at the historical background to some of the key phenomena that shaped the controversy-laden campaign of 2

  6. #6 peak

    2014|⏎|ISBN: ||9781312744967 (DRM-EPUB)|⏎|

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:

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

    873) the Sarmatian ideal is contrasted with Western cultural influences. Rzewuski was the author of the novels Cracow Castle (1847–48; Russian translation, 1875

  2. #2 peak

    The Central Business District as seen from the top of the Carlton Centre⏎|Municipality||City of Johannesburg|⏎|• Total||1.09 km2 (0.42 sq mi)|⏎|• Density||13,0

  3. #3 peak

    which was purchased by the joined States in 1803. subsequent to several territorial reorganizations, Minnesota in its current form was admitted as the country’s 32nd own up on May 11, 1858

  4. #4 peak

    10,000 BCE to 300 BCE), and Haniwa figures, which were used as burial objects during the Kofun period (300 CE to 710 CE). Over time, Japanese

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. L25:146 strongly excites
  2. L25:854 strongly suppresses
  3. L20:787 strongly excites
  4. L21:339 strongly excites spot just before a year
  5. L25:232 strongly suppresses
  6. L22:2275 strongly excites
  7. L25:95 strongly suppresses
  8. L25:182 strongly excites
  9. L22:193 strongly suppresses
  10. L24:2744 strongly suppresses
  11. L20:433 strongly suppresses
  12. L19:1174 strongly excites
  13. L14:140 strongly suppresses
  14. L24:197 excites
  15. L21:395 excites
  16. L23:262 suppresses
  17. L21:613 suppresses
  18. L22:743 suppresses
  19. L24:13 excites
  20. L23:90 suppresses
  21. L23:199 suppresses
  22. L18:246 excites
  23. L16:396 suppresses
  24. L19:430 excites

Feeds later layers

  1. L27:113 strongly suppresses
  2. L27:287 strongly excites
  3. L27:2172 strongly suppresses
  4. L27:306 strongly excites
  5. L27:163 strongly suppresses
  6. L27:448 strongly suppresses
  7. L27:978 excites
  8. L27:1456 excites
  9. L27:1117 suppresses
  10. L27:126 suppresses
  11. L27:962 excites
  12. L27:283 suppresses the spot right before a number
  13. L27:731 suppresses
  14. L27:33 suppresses
  15. L27:428 excites
  16. L27:862 suppresses
  17. L27:462 weakly suppresses
  18. L27:902 weakly suppresses
  19. L27:97 weakly suppresses
  20. L27:1534 weakly excites
  21. L27:2089 weakly excites
  22. L27:1196 weakly suppresses
  23. L27:170 weakly suppresses unclear: spaces in mixed-language text
  24. L27:241 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 -45.75 and -44.021 values between -44.02 and -42.281 values between -40.55 and -38.811 values between -37.08 and -35.351 values between -35.35 and -33.611 values between -30.14 and -28.412 values between -28.41 and -26.683 values between -26.68 and -24.943 values between -24.94 and -23.212 values between -23.21 and -21.472 values between -21.47 and -19.744 values between -19.74 and -18.015 values between -18.01 and -16.2712 values between -16.27 and -14.5412 values between -14.54 and -12.8016 values between -12.80 and -11.0718 values between -11.07 and -9.3424 values between -9.34 and -7.6036 values between -7.60 and -5.8723 values between -5.87 and -4.1345 values between -4.13 and -2.40130 values between -2.40 and -0.673360 values between -0.67 and 1.07145 values between 1.07 and 2.8019 values between 2.80 and 4.547 values between 4.54 and 6.2714 values between 6.27 and 8.007 values between 8.00 and 9.743 values between 9.74 and 11.475 values between 11.47 and 13.212 values between 13.21 and 14.941 values between 14.94 and 16.672 values between 21.88 and 23.61offnegativepositiveoften

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. ivism
  2. XX
  3. →→→→→→
  4. Î
  5. dın
  6. entifier
  7. バランス
  8. の
  9. zero

Suppresses

  1. ⏎ ⏎
  2. :\/\/
  3. ⏎ ⏎
  4. ⏎ ⏎
  5. ⏎ ⏎
  6. -LAST
  7. //**⏎
  8. OPTARG
  9. rending
  10. 十三五