Neuron Explorer

Qwen3-0.6B

the spot before a year number

A neuron in layer 5 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

The neuron fires on the empty spot right before a year, such as after "beginning in" or "born in", when the next word is about to be a year like 1961 or 2009. Turning it up gave rambling text that did not clearly relate to dates.

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

    spacewalk on June 3.⏎Astronauts have been photographing Earth from space since the early Mercury missions beginning in 1961. Today, they observe the Earth from their unique point of

  2. #2 peak

    well known, this hostility between evolution and Christianity continued into the twentieth century, most famously in 1925 in Dayton Tennessee, when the young schoolteacher John Thomas Sc

  3. #3 peak

    “colonizer” on the sculpture’s pedestal.⏎There was also apparently an attempt to topple the statue of the abolitionist:⏎Baldwin, born in 1795, moved to Philadelphia from his birthplace of Elizabet

  4. #4 peak

    Act to let 18-year-olds vote.⏎The Supreme Court, in Oregon v Mitchell, limited this right so the 26th Amendment was passed in 1971 to confirm it.⏎President Nixon stated March 2

  5. #5 peak

    the middle of the last decade and perhaps longer, the growth path and survival rate of new businesses means they are generating fewer new jobs.⏎The businesses that started in 2009, for example, are on course to create 1 million

  6. #6 peak

    shieldmaid<|endoftext|>My class, this article dating from the difference between relative dating - places rock layers to answer to the placement of a fossils found out. In which of absolute age by several. A point. Scientists used in years, or event.W

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: think small, 100% of the time, 100%... I'm a little tired. Let me do it. I'm a little tired, I will have to... I am a little tired. Let me

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

    on this day, January<|endoftext|>The data are in, and the numbers are unequivocal.⏎After all the tumult, after the industry-fostered head-in-sand denialism, after the legislative tinkering to obfuscate the emerging reality,

  2. #2 peak

    limit calculation is considered to be difficult are mentioned here in this section.⏎This is a learning difficulty which is commonly found in students across the globe. In this learning difficulty a student can find trouble in learning and understanding shapes, formulas, and other numeric

  3. #3 peak

    see blood spots on your sheets, live or dead bugs, molted skins, spots of fecal matter, or tiny eggs (usually found in dark crevices).⏎While bed bugs may carry disease, they are not observed to pass disease to

  4. #4 peak

    away run the lizards!⏎Ecologist Sean Doody discovered this while in the field with student Phillip Paull. The two were lifting eggs out of a crevice when . . . Holy Dooley! The hatchlings literally popped out and sprint

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. L3:311 suppresses
  2. L3:265 suppresses
  3. L2:766 suppresses
  4. L4:2209 excites
  5. L2:1073 suppresses
  6. L3:528 weakly excites
  7. L3:369 weakly suppresses
  8. L4:572 weakly excites
  9. L2:461 weakly suppresses
  10. L2:1394 weakly suppresses
  11. L2:488 weakly excites
  12. L4:682 weakly suppresses
  13. L3:520 weakly suppresses
  14. L3:382 weakly suppresses
  15. L4:39 weakly suppresses
  16. L2:658 weakly excites
  17. L3:662 weakly suppresses
  18. L4:362 weakly excites
  19. L0:1913 weakly excites
  20. L0:2338 weakly suppresses
  21. L1:466 weakly excites
  22. L2:491 weakly suppresses
  23. L1:231 weakly suppresses
  24. L4:464 weakly suppresses

Feeds later layers

  1. L6:894 strongly suppresses
  2. L8:359 suppresses
  3. L6:421 suppresses
  4. L10:262 suppresses
  5. L7:1134 suppresses
  6. L8:563 suppresses
  7. L9:2442 suppresses
  8. L9:509 suppresses
  9. L7:1185 suppresses
  10. L11:741 suppresses
  11. L12:1792 suppresses
  12. L8:133 excites
  13. L12:1255 suppresses
  14. L15:853 suppresses
  15. L13:2548 suppresses
  16. L9:361 suppresses
  17. L7:2159 suppresses
  18. L14:1848 suppresses
  19. L10:472 suppresses
  20. L12:631 suppresses
  21. L8:54 excites
  22. L6:695 excites
  23. L12:870 suppresses
  24. L13:921 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 -2.63 and -2.452 values between -1.75 and -1.581 values between -1.58 and -1.403 values between -1.40 and -1.236 values between -1.23 and -1.0516 values between -1.05 and -0.8818 values between -0.88 and -0.7034 values between -0.70 and -0.5366 values between -0.53 and -0.35161 values between -0.35 and -0.181185 values between -0.18 and -0.001793 values between -0.00 and 0.17348 values between 0.17 and 0.35103 values between 0.35 and 0.5246 values between 0.52 and 0.7024 values between 0.70 and 0.8730 values between 0.87 and 1.0511 values between 1.05 and 1.2224 values between 1.22 and 1.403 values between 1.40 and 1.575 values between 1.57 and 1.7410 values between 1.74 and 1.923 values between 1.92 and 2.093 values between 2.09 and 2.273 values between 2.27 and 2.441 values between 2.62 and 2.792 values between 2.79 and 2.974 values between 2.97 and 3.141 values between 3.14 and 3.321 values between 4.19 and 4.37offnegativepositiveoften

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. ước
  2. möglich
  3. đáo
  4. прав
  5. automát
  6. Gäste
  7. adamente
  8. steller
  9. gomery
  10. )&&(

Suppresses

  1. ounds
  2. 面前
  3. 眼里
  4. es
  5. smelled
  6. smells
  7. ES
  8. Describe
  9. unt
  10. -rate