Neuron Explorer

Qwen3-0.6B

connecting word after a full clause

A neuron in layer 19 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on words like 'but', 'though', 'even if', 'because' or a comma that join one full statement to another, especially a contrast or a cause. Turning it up gave text that drifted on topic, but the link is not clear.

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

    , or drilling the item into the property.⏎For example, when a homeowner buys a bathtub at a store, it starts off as personal property. However, the bathtub becomes real property after it

  2. #2 peak

    your toes. This will prevent strain on your knee joints.⏎Burn Calories, but at a Slower Rate⏎You will burn calories using a rowing machine, but not as many as when running. According to the American Council on Exercise, a

  3. #3 peak

    have been mixed. Some clothing companies have paid compensation funds to victims, even though they didn’t employ anyone at Rana Plaza at the time of the accident,

  4. #4 peak

    cache turned on, Drupal stores all the HTML code for any page visited by an anonymous user directly into the database. When another request for the same page comes along, Drupal knows to fetch this page out of the database rather than re-generating it

  5. #5 peak

    had a C-section were less likely to do pelvic floor exercises, probably because they didn’t push the baby out with the pelvic floor. But pregnancy itself that

  6. #6 peak

    , in conjunction with the state Department of Natural Resources, to see how far it's spread.⏎The city could then decide to apply pesticides or cut trees down, but cutting would only be done when the beetle is dormant to prevent further spreading, E

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: its its its / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / /

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

    the risk of developing Alzheimer’s every five years after age 65.⏎Q. What lifestyle changes or other tips do you have to help prevent Alzheimer’s or delay its onset?⏎A. Research has found many lifestyle

  2. #2 peak

    structure in the world – a linear particle accelerator first dubbed Project M and affectionately known as "the Monster" to the scientists who conjured it – to explore the mysterious subatomic realm. Fifty years later, more than

  3. #3 peak

    into an all-engulfing process, why it is so difficult to stop once it has begun, and what our best hopes are for ending war or at least limiting its dangers.⏎Several important developments have taken place since Dr. Dyer published the

  4. #4 peak

    used for ropes is called a sheave. A single sheave mounted in a block and fixed in place simply changes the direction of force exerted on the rope passing over it. If the end of the rope that ordinarily would attach to the load

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. L18:2475 strongly excites
  2. L16:2832 suppresses
  3. L16:1178 excites
  4. L18:358 excites
  5. L18:2698 excites
  6. L17:192 suppresses
  7. L18:2522 suppresses
  8. L18:636 suppresses
  9. L16:2515 excites
  10. L16:577 excites
  11. L17:444 suppresses
  12. L17:1867 suppresses
  13. L18:1220 excites
  14. L18:1504 excites
  15. L17:1536 excites
  16. L18:2470 excites
  17. L15:1226 suppresses
  18. L18:2753 excites
  19. L15:1047 suppresses
  20. L17:491 suppresses
  21. L15:2100 suppresses
  22. L15:2893 excites
  23. L17:1292 excites
  24. L15:3069 excites

Feeds later layers

  1. L23:1001 strongly suppresses
  2. L27:1047 strongly suppresses
  3. L20:2602 strongly suppresses
  4. L22:1029 strongly suppresses
  5. L27:953 strongly suppresses
  6. L20:1229 strongly suppresses
  7. L25:1279 strongly suppresses
  8. L24:2392 suppresses
  9. L23:334 suppresses
  10. L21:1091 suppresses
  11. L24:1127 excites
  12. L24:82 excites
  13. L26:763 suppresses
  14. L23:1203 suppresses
  15. L21:796 suppresses
  16. L26:908 suppresses
  17. L27:1495 suppresses
  18. L27:2345 suppresses
  19. L27:1646 suppresses
  20. L21:1276 excites
  21. L21:2413 suppresses
  22. L25:692 weakly suppresses
  23. L23:974 weakly excites
  24. L24:477 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 -13.17 and -12.652 values between -10.57 and -10.051 values between -9.52 and -9.001 values between -9.00 and -8.481 values between -7.96 and -7.441 values between -6.92 and -6.405 values between -5.88 and -5.352 values between -5.35 and -4.832 values between -4.83 and -4.3112 values between -4.31 and -3.797 values between -3.79 and -3.276 values between -3.27 and -2.7519 values between -2.75 and -2.2325 values between -2.23 and -1.7156 values between -1.71 and -1.18105 values between -1.18 and -0.66379 values between -0.66 and -0.142082 values between -0.14 and 0.381033 values between 0.38 and 0.9090 values between 0.90 and 1.4232 values between 1.42 and 1.9416 values between 1.94 and 2.4610 values between 2.46 and 2.997 values between 2.99 and 3.512 values between 3.51 and 4.032 values between 4.03 and 4.552 values between 4.55 and 5.072 values between 5.07 and 5.592 values between 5.59 and 6.111 values between 6.11 and 6.631 values between 6.63 and 7.161 values between 7.16 and 7.68offnegativepositiveoften

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. -Al
  2. ---------------⏎
  3. evin
  4. حياء
  5. )][
  6. -G
  7. .ordinal
  8. -Un
  9. ----------------------------------------------------------------------⏎
  10. --------------⏎

Suppresses

  1. its
  2. Its
  3. 其
  4. 它的
  5. Its
  6. 对其
  7. 为其
  8. 对其进行
  9. 使其
  10. 其所