Neuron Explorer

Qwen3-0.6B

unclear: numbers and distance words

A neuron in layer 23 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

It mostly lights up on digits inside measurements and on words about how far apart things are, like feet, miles and distance. It is hard to tell what ties these together. Turning it up made the model write about random topics, and turning it down made it repeat the word distance.

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

    (usually Thomas of Cana) with the Thomas; but it is undeniable that the Nasrani or Saint Thomas community in India is very old, going back at least

  2. #2 peak

    . The comet is traveling toward the sun at 34,175 mph. After a high-speed chase across nearly 4 billion miles (6.4 billion kilometers) of interplanetary space Rosetta had to slam on its brakes

  3. #3 peak

    Praying also helps keep us out of trouble. Jesus told us to pray, "Watch and pray so that you will not fall into temptation." (Matthew 26:41)⏎Read James

  4. #4 peak

    and travel parallel to each other (parallel to the axis of symmetry as well).⏎: A parabola is the set of all points in a plane that are equidistant from the focus and the directrix of the parabola. [

  5. #5 peak

    969. Questions or concerns relating to the accessibility of the FCC's online public file system should be directed to the FCC at 888-225-5322, at 888-835

  6. #6 peak

    upgraded to Hurricane Audrey. Air Force reconnaissance investigating Hurricane Audrey over the southwest Gulf of Mexico found a central pressure of 973 millibars or 28.72 inHg.

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: a 2005 a 100005 2007 10012 10025 10040 10050 1005

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

    endurance athletes, and particularly marathon and ultramarathon runners use to keep them fueled during such a long workout:⏎-C

  2. #2 peak

    it will release its spores. The spores are released into the air and can travel many miles to their new home where they will grow through mitosis into a new

  3. #3 peak

    Gulo gulo luscus) is a medium-sized, solitary carnivore adapted for digging, climbing, and traveling long distances in deep snow during the winter. Since the wolverine is a highly

  4. #4 peak

    marathon-mode -I see at least two or three cases of ITBS a day. It’s a very common injury in runners - ones who are trying to do long-distance running

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. L16:959 suppresses
  2. L21:1767 suppresses
  3. L21:2873 excites
  4. L20:1362 excites
  5. L19:2238 suppresses
  6. L22:1886 suppresses
  7. L21:1566 excites
  8. L22:1820 suppresses
  9. L17:2571 excites
  10. L20:2226 suppresses
  11. L22:1000 excites
  12. L20:2627 weakly suppresses
  13. L21:3052 weakly excites
  14. L21:2547 weakly suppresses
  15. L21:235 weakly excites
  16. L4:2856 weakly suppresses
  17. L22:2328 weakly excites
  18. L22:2790 weakly suppresses
  19. L10:2474 weakly suppresses
  20. L21:1177 weakly excites
  21. L17:2308 weakly excites
  22. L17:1511 weakly suppresses
  23. L8:2828 weakly suppresses
  24. L8:918 weakly excites

Feeds later layers

  1. L24:2692 strongly suppresses
  2. L26:2180 strongly suppresses
  3. L27:2830 suppresses
  4. L26:2281 suppresses
  5. L27:1178 suppresses
  6. L26:2903 suppresses
  7. L24:712 suppresses 'long' in 'long-term' and 'long-lasting'
  8. L26:1469 suppresses
  9. L24:587 suppresses
  10. L24:1238 suppresses blank space before a number
  11. L27:1469 weakly suppresses
  12. L24:1219 weakly suppresses
  13. L24:2127 weakly suppresses
  14. L25:319 weakly suppresses
  15. L24:2153 weakly suppresses
  16. L26:1910 weakly suppresses
  17. L24:1336 weakly excites
  18. L25:313 weakly suppresses
  19. L24:1394 weakly suppresses
  20. L27:1555 weakly suppresses
  21. L25:1808 weakly suppresses
  22. L25:1354 weakly suppresses
  23. L24:2174 weakly suppresses
  24. L24:1289 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 -47.38 and -45.921 values between -43.02 and -41.561 values between -34.30 and -32.842 values between -24.13 and -22.671 values between -21.22 and -19.771 values between -19.77 and -18.311 values between -18.31 and -16.863 values between -13.95 and -12.502 values between -12.50 and -11.052 values between -11.05 and -9.597 values between -9.59 and -8.142 values between -8.14 and -6.698 values between -6.69 and -5.2312 values between -5.23 and -3.7812 values between -3.78 and -2.3348 values between -2.33 and -0.883031 values between -0.88 and 0.58767 values between 0.58 and 2.034 values between 2.03 and 3.481 values between 3.48 and 4.941 values between 9.30 and 10.75offnegativepositiveoften

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. redient
  2. �回
  3. xiv
  4. rait
  5. elier
  6. ộc
  7. ören
  8. 内阁
  9. AssemblyVersion
  10. 拈

Suppresses

  1. distance
  2. distances
  3. Distance
  4. distance
  5. Distance
  6. (distance
  7. 距离
  8. distancia
  9. 距離
  10. _distance