Neuron Explorer

Qwen3-0.6B

verb slot after treaty or agreement

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the small word that comes right before a verb, mostly when the subject is a treaty, convention or agreement, as in 'the treaty was' before 'signed'. Turning it up made the model write 'sign' and 'signal' words over and over, which fits the idea that it is preparing the word 'signed'.

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

    want to communicate with. The private part is for the recipient's use only.⏎When you send personal information to dell.com, you use Dell's public key to encrypt your personal information. That means, if at any point during the transmission your

  2. #2 peak

    the Bali Declaration.⏎About Minamata Convention on Mercury⏎- The Minamata Convention on Mercury is the most recent global agreement on environment and health. It was adopted in 2013 and entered into force August 16,

  3. #3 peak

    818 Adams and President James Monroe resumed efforts to acquire Florida and a western boundary for the Louisiana Purchase. After months of negotiations, the Adams-Onís Treaty was signed on February 22, 18

  4. #4 peak

    his city's role in the Trans-Atlantic slave trade. He acknowledged that London was "tainted" by it<|endoftext|>Nearly three months after the Treaty of Paris was signed ending the American Revolution, the last British soldiers withdraw from New York City,

  5. #5 peak

    agreed that Texas was on the Spanish side of the line, and Spain agreed to give up its claim to the Northwest Territory north of forty-two degrees. The treaty was approved by the U.S. Senate on February 24, 18

  6. #6 peak

    housing, health and educational needs as well as economic, social and cultural rights.⏎The United Nations Convention on the Rights of the Child, to which New Zealand is a signatory, says that social services is also about ensuring that in all that we

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: the hard working hard working hard working hard working hard working hard working hard working hard working hard working hard working hard working hard working hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard hard

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

    build new nature-inspired devices. Think: Cat-tongue-inspired hairbrush, flexible arms to pick up fruit, and robots that walk on water.⏎Rex Walheim has some

  2. #2 peak

    to do with the success I have enjoyed as a man,” Peterson said in a statement.⏎Among those coming to Peterson’s defense was NBA Hall of Famer Charles Barkley. On the CBS pregame show The NFL Today, Barkley recalled childhood

  3. #3 peak

    , some of the vertebrate tongues, and the elephant trunk. These structures are composed solely of muscles used to generate movement and provide the necessary skeletal support (Kier, 1982; Smith and Kier, 198

  4. #4 peak

    , as well as the residence of his official mistress, Anne, Duchess of Étampes. Francis appears as the patron of Benvenuto Cellini in the 1843 French novel L'Orfèvre

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. L24:1449 excites
  2. L23:2320 excites
  3. L24:1188 suppresses
  4. L23:2878 suppresses
  5. L22:1839 excites
  6. L24:1706 excites
  7. L16:56 weakly suppresses
  8. L2:1142 weakly excites
  9. L18:221 weakly excites
  10. L9:1628 weakly excites
  11. L18:191 weakly suppresses
  12. L19:1630 weakly excites
  13. L25:2254 weakly excites
  14. L18:2554 weakly excites
  15. L10:10 weakly excites
  16. L25:1526 weakly suppresses
  17. L4:871 weakly excites
  18. L25:251 weakly suppresses
  19. L19:2726 weakly suppresses
  20. L21:2327 weakly suppresses
  21. L14:246 weakly suppresses
  22. L0:2122 weakly suppresses
  23. L8:1274 weakly suppresses
  24. L21:2590 weakly suppresses

Feeds later layers

  1. L27:1453 strongly excites
  2. L27:1819 weakly excites
  3. L27:2107 weakly excites
  4. L27:2261 weakly suppresses
  5. L27:522 weakly suppresses
  6. L27:2285 weakly suppresses
  7. L27:2402 weakly excites
  8. L27:3037 weakly suppresses
  9. L27:2924 weakly suppresses
  10. L27:1867 weakly suppresses
  11. L27:2926 weakly excites
  12. L27:653 weakly excites
  13. L27:298 weakly suppresses
  14. L27:2082 weakly excites
  15. L27:827 weakly excites
  16. L27:2032 weakly suppresses
  17. L27:1787 weakly excites
  18. L27:2357 weakly suppresses
  19. L27:654 weakly suppresses
  20. L27:1762 weakly excites
  21. L27:1640 weakly excites
  22. L27:876 weakly suppresses
  23. L27:1767 weakly suppresses
  24. L27:1882 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 -19.72 and -18.511 values between -16.09 and -14.883 values between -13.67 and -12.461 values between -12.46 and -11.252 values between -11.25 and -10.041 values between -10.04 and -8.834 values between -8.83 and -7.621 values between -7.62 and -6.413 values between -6.41 and -5.206 values between -5.20 and -3.997 values between -3.99 and -2.7834 values between -2.78 and -1.57402 values between -1.57 and -0.363228 values between -0.36 and 0.85132 values between 0.85 and 2.0632 values between 2.06 and 3.2719 values between 3.27 and 4.489 values between 4.48 and 5.696 values between 5.69 and 6.904 values between 6.90 and 8.116 values between 8.11 and 9.323 values between 9.32 and 10.531 values between 10.53 and 11.741 values between 18.99 and 20.201 values between 27.46 and 28.67offnegativepositiveoften

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. sign
  2. signs
  3. sign
  4. Sign
  5. Sign
  6. .sign
  7. (sign
  8. Signs
  9. _sign
  10. SIGN

Suppresses

  1. leness
  2. .tencent
  3. IEWS
  4. ollections
  5. .defineProperty
  6. DispatchQueue
  7. .TestCase
  8. 因地制宜
  9. erry
  10. entityManager