Neuron Explorer

Qwen3-0.6B

unclear: document ends and timestamps

A neuron in layer 14 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 strongest cases are a mix: the end-of-document marker, and spots inside date and time stamps. The negative cases are pieces of words split in the middle of a name. We could not find one clear story for this neuron, and the steered outputs do not point to a single theme.

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

    will print you need to store it in 6 bytes, with the last byte equal to 0.⏎October 3rd, 2010, 07:44 AM⏎We are not going to do you homework.⏎

  2. #2 peak

    other early math activities.⏎Do you have any other fun ways that you re-use candy at your house?⏎What else have you made from an egg carton?<|endoftext|>Newport Aquarium's 150-year-old turtle dies⏎Newport Aquarium

  3. #3 peak

    support your answer.⏎problem6. Compare and contrast a peer-to-peer network architecture to Client-Server architecture. Support your answer with proper diagrams and exs.<|endoftext|>The Minister for Housing, Local Government and Heritage, Darragh O’Brien,

  4. #4 peak

    being caused by curve space when neither were in motion, what causes the force to act upon each other.⏎Hope this is clear enough to get my question across.<|endoftext|>Yesterday I cut myself some fresh pineapple, then promptly proceeded to

  5. #5 peak

    al-install, you can manually download and install bool-extras:⏎tar xzf bool-extras-0.3.0.tgz⏎./Setup.lhs<|endoftext|>Education Pioneers recruits, develops, and connects professionals from diverse backgrounds to resolve

  6. #6 peak

    taste works⏎Scientists have discovered an ion channel that allows taste buds to communicate with the brain.⏎Wed, Mar 06, 2013 at 03:20 PM⏎taste buds on mouse tongue (Photo:

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 need to find the number of the number of the number of the number of the number of the number of the number of this is the number of this is the number of this, and the number of this, where the number is a

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

    of which are particularly flexible, while others carry meshes of wires within them for electrical connectivity, and yet others act as insulators.<|endoftext|>Heb. 'eshel ( Genesis 21:33 ; 1 Samuel 22:

  2. #2 peak

    ., in association with Lou Weller (Caddo), the Native American Design Collaborative, and Polshek Partnership Architects of New York City; Ramona Sakiestewa (Hopi) and Donna House (Navajo/Oneida)

  3. #3 peak

    from 20 countries in a Bangkok, Thailand, workshop to help set up a tsunami early warning system for the Indian Ocean, in a bid to avoid a reply of the December 2004 tsunami. Presumably, many of

  4. #4 peak

    indoor evaporator coil. The heat laden freon then flows back out to the outdoor condensing unit, which then dispells the heat to the outside, via the outdoor condenser coil.⏎Both the indoor

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. L11:262 strongly excites
  2. L11:321 strongly excites
  3. L11:435 strongly suppresses
  4. L11:232 strongly suppresses
  5. L13:415 strongly suppresses
  6. L11:235 strongly suppresses
  7. L13:37 strongly suppresses
  8. L11:183 strongly excites
  9. L13:183 strongly suppresses
  10. L13:87 strongly suppresses
  11. L13:65 strongly suppresses
  12. L12:263 strongly excites
  13. L13:117 excites
  14. L13:256 suppresses
  15. L12:323 excites
  16. L11:284 suppresses
  17. L13:343 suppresses
  18. L13:219 suppresses
  19. L12:209 excites
  20. L12:338 suppresses
  21. L13:253 suppresses
  22. L12:121 excites
  23. L11:2429 excites
  24. L13:177 suppresses

Feeds later layers

  1. L17:826 excites
  2. L15:2313 suppresses
  3. L15:1627 excites
  4. L17:2271 suppresses
  5. L15:431 suppresses
  6. L15:538 excites
  7. L16:2114 suppresses
  8. L15:1251 excites
  9. L16:1116 excites
  10. L17:748 suppresses
  11. L16:596 suppresses
  12. L15:453 suppresses
  13. L16:851 excites
  14. L16:870 excites
  15. L15:375 suppresses
  16. L15:1076 suppresses
  17. L15:1071 excites
  18. L18:214 suppresses end of document marker
  19. L15:472 suppresses
  20. L16:1874 excites
  21. L15:1186 suppresses
  22. L16:1974 excites
  23. L17:614 suppresses
  24. L16:1187 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 -3.70 and -3.542 values between -3.04 and -2.882 values between -2.72 and -2.554 values between -2.55 and -2.391 values between -2.39 and -2.226 values between -2.22 and -2.064 values between -2.06 and -1.909 values between -1.90 and -1.7312 values between -1.73 and -1.5717 values between -1.57 and -1.4027 values between -1.40 and -1.2448 values between -1.24 and -1.0856 values between -1.08 and -0.91100 values between -0.91 and -0.75120 values between -0.75 and -0.59186 values between -0.59 and -0.42306 values between -0.42 and -0.26547 values between -0.26 and -0.091046 values between -0.09 and 0.07583 values between 0.07 and 0.24400 values between 0.24 and 0.40245 values between 0.40 and 0.5685 values between 0.56 and 0.7346 values between 0.73 and 0.8918 values between 0.89 and 1.0515 values between 1.05 and 1.226 values between 1.22 and 1.383 values between 1.38 and 1.554 values between 1.55 and 1.714 values between 1.71 and 1.871 values between 1.87 and 2.041 values between 2.04 and 2.201 values between 2.20 and 2.371 values between 2.37 and 2.531 values between 2.69 and 2.86offnegativepositiveoften

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. ponsible
  2. ista
  3. naire
  4. Powered
  5. 骄傲
  6. 的背后
  7. ่น
  8. lijk
  9. 四级
  10. 中新

Suppresses

  1. ieg
  2. еры
  3. ете
  4. AML
  5. amel
  6. cá
  7. emas
  8. flen
  9. amus
  10. ums