Neuron Explorer

Qwen3-0.6B

'rec' starting re-verbs like recover

A neuron in layer 24 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 piece 'rec' at the start of words like 'recover', 'recite' and 'recede', where it is a prefix meaning 'back' or 'again'. Turning it up or down did not give steered text with a clear link to that pattern.

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

    , NOAA's Deke Arndt, Chief of the Climate Monitoring Branch at the National Climatic Data Center, recaps the temperature and precipitation data for the continental US in summer 201

  2. #2 peak

    around the North Bay," says Jenniskens. "Much more accurate results will follow from a comprehensive study of the video records. Now, we hope that someone recovers a meteorite on the ground."⏎In the map, red dots represent the

  3. #3 peak

    helping kids apply knowledge, think critically, imagine, be creative, be confident about being lifelong learners. It’s not simply about memorising facts and being able to recite that which one has memorised; it’s about knowing how to learn and

  4. #4 peak

    Head of the Oceanography Department at Texas A&M University says that if a storm surge were to move inland, the oil would be left behind even after the waters recede. The oil has a tendency to stick to organic matter, and there would

  5. #5 peak

    of helping or doing work for someone”. Does a microservice fit into this general definition? In order to come up with a more definitive answer, lets recollect the knowledge that is already out there.⏎“Microservices aim to do SO

  6. #6 peak

    us, the treatment of obstructive bronchitis in children in the home, it is directed to the liquefaction and excretion out.As a result, it recovers patency respiratory system.⏎to reduce spasms in the bronchi can be

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 want to get more support for the IICIC. I see that IICIC is part of the IICIC series, which I also run, and I need to get more support for IICIC. I also run the

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

    compartment. If this occurs, the freezer may still get cold, while the refrigerator will not get cold. To determine if the evaporator fan motor is defective, try turning the fan blade by hand. If the fan blade does not turn freely,

  2. #2 peak

    is not working properly, the refrigerator won’t cool properly. To determine if the fan motor is defective, first check the fan blade for obstructions. Next, try turning the fan motor blade by hand. If the blade

  3. #3 peak

    to math needed<|endoftext|>A screw conveyor is a mechanism that uses a rotating helical screw, called “flighting”, that is installed within a tube or trough, to move liquid or granular materials. They are used in many bulk handling industries.⏎

  4. #4 peak

    sand dollar each time⏎you choose the correct answer. Keep track of how many sand dollars<|endoftext|>Dredging is referring to those engineering measures using dredger equipment to excavate those underwater soil and stones to increase the depth of water and remove the

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. L19:730 excites
  2. L19:1626 suppresses
  3. L22:181 suppresses
  4. L23:53 excites
  5. L16:1823 suppresses
  6. L21:239 suppresses
  7. L18:2131 suppresses
  8. L21:943 excites
  9. L20:1124 suppresses
  10. L6:2359 excites
  11. L23:2922 suppresses
  12. L19:2932 suppresses
  13. L19:2879 weakly suppresses
  14. L19:2632 weakly suppresses
  15. L2:2384 weakly suppresses
  16. L16:1153 weakly suppresses
  17. L4:17 weakly suppresses
  18. L14:659 weakly excites
  19. L13:909 weakly excites 'to' before a verb after a noun
  20. L6:1133 weakly excites "to" before an infinitive verb
  21. L22:548 weakly excites
  22. L22:2323 weakly excites
  23. L1:870 weakly suppresses
  24. L20:1561 weakly excites

Feeds later layers

  1. L25:2793 weakly excites
  2. L25:2176 weakly suppresses
  3. L27:1929 weakly suppresses
  4. L25:2769 weakly excites
  5. L27:1958 weakly suppresses
  6. L27:2796 weakly excites
  7. L27:1216 weakly suppresses
  8. L26:2240 weakly suppresses
  9. L27:2903 weakly excites
  10. L26:435 weakly excites
  11. L25:3022 weakly excites
  12. L26:1287 weakly excites
  13. L26:1659 weakly excites
  14. L25:1490 weakly suppresses
  15. L26:1724 weakly suppresses
  16. L27:2686 weakly excites
  17. L26:3056 weakly suppresses
  18. L27:210 weakly suppresses
  19. L27:496 weakly suppresses
  20. L25:2664 weakly suppresses
  21. L26:1655 weakly excites
  22. L26:880 weakly excites
  23. L27:2057 weakly suppresses
  24. L25:1497 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 -27.77 and -26.741 values between -15.44 and -14.411 values between -14.41 and -13.381 values between -13.38 and -12.362 values between -12.36 and -11.335 values between -10.30 and -9.274 values between -9.27 and -8.252 values between -8.25 and -7.222 values between -7.22 and -6.198 values between -6.19 and -5.1615 values between -5.16 and -4.1412 values between -4.14 and -3.1128 values between -3.11 and -2.0887 values between -2.08 and -1.061882 values between -1.06 and -0.031485 values between -0.03 and 1.00179 values between 1.00 and 2.0371 values between 2.03 and 3.0645 values between 3.06 and 4.0831 values between 4.08 and 5.1121 values between 5.11 and 6.1413 values between 6.14 and 7.164 values between 7.16 and 8.194 values between 8.19 and 9.222 values between 9.22 and 10.251 values between 11.27 and 12.301 values between 12.30 and 13.33offnegativepositiveoften

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. brings
  2. bringing
  3. bring
  4. bring
  5. 赋予
  6. bracht
  7. させ
  8. 给他们
  9. ่า
  10. generates

Suppresses

  1. register
  2. spoilers
  3. 蜢
  4. зв
  5. registering
  6. 穴
  7. spe
  8. раж
  9. Track
  10. runway