Neuron Explorer

Qwen3-0.6B

ordinal numbers in military unit names

A neuron in layer 9 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 "th" in numbered military units, like the 39th Fighter Squadron or the 84th Indiana Infantry. When it was turned up, the model's writing drifted toward soldiers, armies and the military.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    -86 Sabres took place in the skies over North Korea. Captain Joseph C. McConnell, Jr., leading a flight of fighters of the 39th Fighter Interceptor Squadron, 51st Fighter Interceptor Wing, based at

  2. #2 peak

    began to recruit a second all-black regiment of soldiers to join the fight.⏎As the struggle against the Confederate South raged on, the newly-formed 54th Massachusetts battalion prepared to attack the seemingly impregnable fortress of Fort Wagner. The

  3. #3 peak

    Army. Indiana Infantry Regiment, 84th (1862-1865)⏎Historical Note: The

  4. #4 peak

    May 14, 1860.⏎With the coming of the Civil War, Tinkham helped to organize Company F of the 26th Illinois Volunteers regiment and was commissioned as Captain on August 10, 1

  5. #5 peak

    allergic reactions, you should talk with your doctor and request a vaccine with the least egg protein available<|endoftext|>United States. Army. Indiana Infantry Regiment, 84th (1862-1865) | Purdue University Libraries,

  6. #6 peak

    of New Alexandria to learn the blacksmith trade before enlisting in the Union Army.⏎In 1861, William enlisted in the Pennsylvania 11th Reserve Infantry. He was just 18 and small in stature. Military officials

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: when I look at the graph of a function, say, a line, I can see that the function is a curve that crosses the x-axis. If I use a calculator to find the value of the product of two numbers, say,

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

    . An experimental 500-w laser built for the U.S. Army was field-tested successfully, destroying 98% of land mines and unexploded mortar shells in only a few minutes. The laser ignited the explosives by heating their

  2. #2 peak

    costs about $148 million.⏎Needless to say, the F-22 is kind of a big deal. Going head to head with the multi-role fighter under any circumstance sounds like a bad idea, given that its primary function is

  3. #3 peak

    Users, 2012-2017). According to Foreign Policy, The Defence Advanced Research Projects Agency is developing some extremely high technology to tackle possible alien invasions, an example being

  4. #4 peak

    ments, particularly following Hitler's infamous ‘Commando Order’ issued on 18 October 1942, ordering the annihilation of all men operating on Commando raids against

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. L8:2626 excites
  2. L6:2995 excites
  3. L7:2377 weakly suppresses
  4. L5:133 weakly excites
  5. L6:873 weakly excites
  6. L6:2332 weakly excites
  7. L8:10 weakly excites
  8. L3:2294 weakly excites
  9. L4:2923 weakly suppresses
  10. L2:554 weakly suppresses
  11. L5:2085 weakly suppresses
  12. L5:2135 weakly excites
  13. L6:1608 weakly excites
  14. L7:1513 weakly excites
  15. L7:2379 weakly suppresses
  16. L7:2366 weakly suppresses
  17. L8:89 weakly excites
  18. L4:1700 weakly excites conservation laws and invariance
  19. L5:495 weakly suppresses
  20. L2:3070 weakly excites
  21. L5:57 weakly suppresses
  22. L3:916 weakly excites
  23. L6:1459 weakly excites
  24. L6:1774 weakly excites

Feeds later layers

  1. L18:2097 excites
  2. L20:2387 excites
  3. L24:2720 excites
  4. L19:2937 excites unclear: mixed possessives and fragments
  5. L16:1460 excites 'of' after a group noun
  6. L11:1519 excites
  7. L19:2473 excites
  8. L12:1348 suppresses
  9. L22:1196 excites
  10. L17:978 excites
  11. L21:372 weakly excites
  12. L10:2608 weakly excites
  13. L10:2261 weakly suppresses
  14. L16:2518 weakly excites
  15. L21:3001 weakly excites
  16. L10:1880 weakly excites
  17. L21:1831 weakly excites
  18. L11:1941 weakly excites
  19. L19:2550 weakly suppresses
  20. L23:1669 weakly suppresses
  21. L17:2150 weakly suppresses
  22. L24:2772 weakly suppresses
  23. L20:1247 weakly excites
  24. L18:2997 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 -1.61 and -1.381 values between -1.38 and -1.152 values between -0.92 and -0.6916 values between -0.69 and -0.46315 values between -0.46 and -0.232217 values between -0.23 and -0.001234 values between -0.00 and 0.2387 values between 0.23 and 0.4616 values between 0.46 and 0.693 values between 0.69 and 0.924 values between 0.92 and 1.152 values between 1.15 and 1.382 values between 1.38 and 1.611 values between 1.61 and 1.841 values between 2.07 and 2.301 values between 2.76 and 2.992 values between 2.99 and 3.221 values between 3.22 and 3.452 values between 7.36 and 7.59offnegativepositiveoften

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. 队伍
  2. 队
  3. 军队
  4. Battalion
  5. 班组
  6. crew
  7. Milit
  8. 部队
  9. illery
  10. military

Suppresses

  1. egg
  2. emark
  3. nues
  4. GetProperty
  5. emodel
  6. bart
  7. [item
  8. egot
  9. ULATOR
  10. 妣