Neuron Explorer

Qwen3-0.6B

the word 'local'

A neuron in layer 25 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 fires on the word 'local', as in 'local area network' or 'local government'. Turning it down made the text wander into bridges and predicted values, and turning it up made it produce number lists, so the steering shows no clear link to the word itself.

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

    , local government and military organizations use the networks for communication purposes. In education, we have

  2. #2 peak

    : local and sustainable. However, no new laws on food security will be passed and the

  3. #3 peak

    needs of organizations. Understand how to design networks by using manual or hand calculations. Define the differences between protocols, software, and network architecture. Define the concept of local area networks and describe their use. Describe how a local area network is installed,

  4. #4 peak

    of the new galaxies are gravitationally bound to the Milky Way, while the eighth appears to float freely in space, beyond our galaxy's grasp.⏎The new Local Group members are even smaller and fainter than other known dwarf galaxies, with lumin

  5. #5 peak

    from local traffic, a flutter in electrical current—can plunge the results of the most carefully

  6. #6 peak

    September 13 local wind shear was having its effect on the storm, and Hurricane Humberto’s winds

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 ways in which the passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive passive

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 the Afro-Asiatic language family that includes Hebrew, Arabic, Aramaic, and Egyptian. Of course the African Diaspora would assume there was

  2. #2 peak

    days the Polynesian Seismic Network (RSP for Réseau Sismique Polynésien) recorded 423 small- and medium-amplitude T

  3. #3 peak

    one is to have a separate set of wires known as “sense” wires. They detect the voltage directly at the device without drawing much current (likely by utilizing high input impedance op amps). So the current that is powering the remote device will be

  4. #4 peak

    as cardiovascular medications, are effective in the treatment of tics and also in decreasing anxiety. These medications may be an acceptable first option in some patients.⏎Clonazepam (Klonopin) belongs to a group of medications (the benz

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. L22:2008 strongly excites
  2. L20:1288 excites
  3. L23:2381 suppresses slot before a grade or age number
  4. L24:993 excites
  5. L24:1120 excites
  6. L20:2498 excites
  7. L9:2962 excites
  8. L23:2630 suppresses blank spot where a clock time comes next
  9. L19:2625 suppresses
  10. L9:2539 excites
  11. L17:1294 excites
  12. L10:2020 excites
  13. L17:1785 excites
  14. L20:2685 weakly excites
  15. L17:2633 weakly suppresses
  16. L4:2992 weakly excites
  17. L7:705 weakly suppresses
  18. L21:2158 weakly suppresses
  19. L9:810 weakly suppresses
  20. L23:1023 weakly suppresses
  21. L21:2538 weakly suppresses
  22. L7:3048 weakly suppresses
  23. L21:2172 weakly excites
  24. L3:1783 weakly excites

Feeds later layers

  1. L27:1288 strongly suppresses
  2. L27:2468 suppresses
  3. L27:2894 weakly suppresses
  4. L27:393 weakly suppresses
  5. L27:1625 weakly suppresses
  6. L26:2486 weakly suppresses
  7. L26:2623 weakly suppresses
  8. L27:2412 weakly excites
  9. L26:2425 weakly excites
  10. L26:1591 weakly suppresses
  11. L26:2870 weakly suppresses
  12. L26:2769 weakly suppresses the 'inter' word beginning
  13. L26:1232 weakly suppresses
  14. L26:948 weakly suppresses
  15. L26:2793 weakly suppresses
  16. L26:1189 weakly suppresses
  17. L27:1748 weakly excites
  18. L26:616 weakly suppresses
  19. L27:2536 weakly suppresses
  20. L27:2262 weakly suppresses
  21. L26:1089 weakly excites
  22. L26:1939 weakly suppresses
  23. L26:1059 weakly suppresses
  24. L27:1201 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.

3 values between -10.43 and -7.148 values between -7.14 and -3.85222 values between -3.85 and -0.563562 values between -0.56 and 2.7378 values between 2.73 and 6.0225 values between 6.02 and 9.317 values between 9.31 and 12.602 values between 12.60 and 15.891 values between 117.90 and 121.19offnegativepositiveoften

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. ơ
  2. ANTITY
  3. erator
  4. одерж
  5. flen
  6. тан
  7. ери
  8. apart
  9. ្�
  10. FML

Suppresses

  1. local
  2. (local
  3. local
  4. Local
  5. Local
  6. 本地
  7. →local
  8. locally
  9. .local
  10. -local