Neuron Explorer

Qwen3-0.6B

unclear: word before a clause continues

A neuron in layer 12 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

This neuron fires on a word that sits just before the rest of a sentence finishes, such as "not [because]", "depends on [where]" or "acts committed [with]". The pattern is hard to pin down, and turning it up did not produce any clear 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

    Kennedy at Rice University in Houston in September of 1962:⏎“We choose to go to the Moon in this decade and do the other things, not because they are easy, but because they are hard, because

  2. #2 peak

    help reverse or prevent a brain herniation, the medical team will treat increased swelling and pressure in the brain. Treatment may involve:⏎The outlook varies and depends on where in the brain the herniation occurrs. Without treatment, death is likely.⏎

  3. #3 peak

    kin’s campaign was crowned with success in December 1948, when the United Nations adopted the Genocide Convention. It defined genocide as “acts committed with intent to destroy,<|endoftext|>1905 - 1938⏎

  4. #4 peak

    site, now owned by Agfa-Gevaert, the Belgian photographic film giant, is closed to the public. It is undergoing a state-ordered cleanup for soil and groundwater contamination linked to Peerless Photo Products, a previous owner that was acquired

  5. #5 peak

    it was held, the names of the people who attended and were absent (apologies), the main issues discussed should then be listed and what actions needs to take place. These can be printed off in paper for meetings and can be distributed electronically

  6. #6 peak

    school for children imprisoned with their mothers, and a<|endoftext|>Eating dark chocolate may ease depressive symptoms and have a positive effect on mood, according to a new study led by University College London (UCL).⏎The study, published in the journal Depression

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've been thinking about... (a) I have been thinking about... (a) [a] [until] [until] [until] (a) [until] [until] (a) [until] [until

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

    to run continuously.⏎Reduce indoor air pollution sources such as smoking, burning incense, and frying foods.⏎If travelling in a car with air conditioning, keep the windows up and the ventilation set to recirculate.⏎If you are very sensitive to

  2. #2 peak

    electronic controls are required.⏎Figure 1: Structure of motors.⏎Types of stepping motors⏎There are several basic types of stepping motors:⏎- Variable reluctance motors with metal teeth.⏎- Permanent magnet motors.⏎- Hybrid motors with both permanent magnets and metal

  3. #3 peak

    directions. While returning from one of his campaigns, he was killed by one of his sons, who seized the throne. Bactrian coins were made to the Greek Standard, and this is one of the most beautiful coins of the late Hellen

  4. #4 peak

    to have ever lived. He was the Golden Warrior of the Bronze Age. Even Alexander the Great admired and emulated Achilles throughout his life, sleeping with a copy of the Iliad under his pillow. And yes, for me, Achilles was

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. L10:456 excites
  2. L8:1916 suppresses
  3. L9:783 excites
  4. L8:2545 suppresses
  5. L7:1708 suppresses
  6. L9:306 excites
  7. L11:1669 excites
  8. L11:1349 suppresses
  9. L7:716 excites
  10. L11:1352 suppresses
  11. L7:1616 excites
  12. L11:1030 suppresses
  13. L9:1135 excites
  14. L10:1851 suppresses
  15. L9:1409 suppresses
  16. L11:2586 excites
  17. L11:599 excites
  18. L6:2369 excites
  19. L6:1360 suppresses
  20. L9:1726 weakly suppresses
  21. L8:1749 weakly excites
  22. L11:834 weakly excites
  23. L11:2065 weakly excites
  24. L10:278 weakly excites

Feeds later layers

  1. L13:1824 strongly excites
  2. L15:1731 strongly excites
  3. L18:2260 excites
  4. L13:2615 excites
  5. L13:2213 suppresses
  6. L13:848 suppresses
  7. L14:1514 excites
  8. L13:2866 excites
  9. L18:1319 excites
  10. L13:640 suppresses
  11. L16:2979 excites
  12. L18:1837 excites
  13. L13:3017 suppresses
  14. L13:623 excites
  15. L13:1122 suppresses
  16. L16:2573 excites
  17. L13:2113 suppresses
  18. L17:1432 excites
  19. L13:1028 suppresses
  20. L16:1277 excites
  21. L13:1581 excites
  22. L13:1562 excites unclear: words that set up a qualifier
  23. L15:402 suppresses
  24. L13:506 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 -2.47 and -2.351 values between -2.35 and -2.231 values between -1.99 and -1.874 values between -1.87 and -1.753 values between -1.75 and -1.634 values between -1.63 and -1.519 values between -1.51 and -1.3928 values between -1.39 and -1.2729 values between -1.27 and -1.1532 values between -1.15 and -1.0357 values between -1.03 and -0.9182 values between -0.91 and -0.79111 values between -0.79 and -0.67173 values between -0.67 and -0.55241 values between -0.55 and -0.43316 values between -0.43 and -0.31439 values between -0.31 and -0.19515 values between -0.19 and -0.07555 values between -0.07 and 0.05378 values between 0.05 and 0.17261 values between 0.17 and 0.29214 values between 0.29 and 0.41143 values between 0.41 and 0.5384 values between 0.53 and 0.6568 values between 0.65 and 0.7749 values between 0.77 and 0.8932 values between 0.89 and 1.0126 values between 1.01 and 1.1316 values between 1.13 and 1.255 values between 1.25 and 1.3712 values between 1.37 and 1.492 values between 1.49 and 1.616 values between 1.61 and 1.734 values between 1.73 and 1.851 values between 1.85 and 1.972 values between 1.97 and 2.091 values between 2.09 and 2.213 values between 2.21 and 2.33offnegativepositiveoften

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. 隨時
  2. 随时
  3. 时时
  4. 每月
  5. 每当
  6. 每周
  7. 每次
  8. 每个月
  9. 关键时刻
  10. 问我

Suppresses

  1. asty
  2. older
  3. urate
  4. _old
  5. old
  6. .named
  7. lymp
  8. ridged
  9. igious
  10. elder