Neuron Explorer

Qwen3-0.6B

mid-word pieces of technical words

A neuron in layer 4 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 the middle chunk of long, technical words that get split up, such as 'hed-on-ist' or 'heterozy-g-ous' in genetics. Turning it up or down did not give a clear topic in the writing.

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

    of new maternity: Her black-haired infant is wrapped in a pink fleece blanket and planted at her breast.⏎Her summary makes the cuarentena sound like a hedonist’s dream, until she elaborates: food, sex, and rest are

  2. #2 peak

    25% chance that their offspring would have this autosomal recessive trait or disorder.⏎So let's see what would happen if one parent was heterozygous and one parent was homozygous recess<|endoftext|>How do you get HIV

  3. #3 peak

    combed or spongy appearance. The bone matrix, or framework, is organized into a three-dimensional latticework of bony processes, called trabeculae, arranged along lines of stress. The spaces between are often filled with

  4. #4 peak

    life: rural society comes to represent family, religion, morality and stability, while the chaotic urban life that Paton describes represents the breaking up of families, hedonism, and atheism. Paton also illustrates this theme through the development of

  5. #5 peak

    particular, that the crystals were pure protein, and assembled by autocatalysis⏎) were incorrect. The first electron microscopical images of TMV were made in 1939 by Gustav

  6. #6 peak

    in many parts of Tibet, with shops as well, where they were not seen previously.⏎Tibet's Overall Population⏎U'Tsang (Tibet Autonomous Region)⏎In the capital Lhasa, Tibetans offficially

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 way I have been thinking about, the way I have been thinking about, the way I have been thinking about... I mean, I just need to be sure I know all the way through that. The way I have been thinking about

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

    each human.”⏎This year’s ceremony was more lavish than the one Rendeh witnessed nearly half a century ago.⏎“That last one was not so large as this. We were poorer,” said R

  2. #2 peak

    we think would be optimized for military use, particularly for forward operating bases. We’re also trying to design new solar cells that would be particularly advantageous to the military. And our fourth task is to set up a test

  3. #3 peak

    unthought of in the South," he said. "People like William Holmes Borders, Daddy King [Martin Luther King Sr.], John C. Calhoun. Those were the people who sort of set Atlanta on the course for becoming a very

  4. #4 peak

    at UBC said, “Sepsis triggers an uncontrolled chain-reaction of inflammation in the body, leading to tissue damage, organ failure, and death. We have discovered that the enzyme ABCF1 acts as a

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. L3:1223 excites
  2. L0:1403 weakly excites
  3. L1:1323 weakly excites
  4. L2:2344 weakly excites
  5. L2:2365 weakly suppresses
  6. L3:626 weakly excites
  7. L1:1130 weakly suppresses
  8. L2:489 weakly suppresses
  9. L2:2057 weakly excites
  10. L0:1095 weakly excites
  11. L1:2317 weakly suppresses
  12. L1:1074 weakly excites
  13. L2:549 weakly suppresses
  14. L3:1335 weakly suppresses
  15. L3:400 weakly excites
  16. L3:2987 weakly excites
  17. L1:1014 weakly excites
  18. L3:2567 weakly excites
  19. L3:1644 weakly excites
  20. L1:1731 weakly suppresses
  21. L2:1470 weakly excites
  22. L0:1707 weakly suppresses
  23. L0:549 weakly excites
  24. L2:2382 weakly suppresses

Feeds later layers

  1. L5:506 excites the word tibia (shin bone)
  2. L10:94 excites
  3. L24:118 excites
  4. L6:853 excites
  5. L6:177 suppresses
  6. L16:1231 excites
  7. L17:446 excites
  8. L22:271 suppresses
  9. L22:1055 excites
  10. L21:327 excites
  11. L19:395 excites
  12. L21:625 excites
  13. L16:286 excites
  14. L19:1204 excites
  15. L18:1399 excites
  16. L6:1100 excites
  17. L15:2378 excites
  18. L25:1493 excites
  19. L26:402 excites
  20. L18:853 suppresses
  21. L9:815 excites
  22. L23:806 excites
  23. L26:2496 excites
  24. L7:2975 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 -2.78 and -2.681 values between -1.50 and -1.401 values between -1.30 and -1.201 values between -1.20 and -1.102 values between -1.10 and -1.001 values between -1.00 and -0.912 values between -0.91 and -0.814 values between -0.81 and -0.713 values between -0.71 and -0.614 values between -0.61 and -0.517 values between -0.51 and -0.4126 values between -0.41 and -0.3181 values between -0.31 and -0.22231 values between -0.22 and -0.12757 values between -0.12 and -0.021418 values between -0.02 and 0.08927 values between 0.08 and 0.18349 values between 0.18 and 0.2872 values between 0.28 and 0.3815 values between 0.38 and 0.481 values between 0.48 and 0.571 values between 0.57 and 0.672 values between 0.67 and 0.771 values between 1.07 and 1.17offnegativepositiveoften

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. ️
  2. ature
  3. ers
  4. 者
  5. ся
  6. ation
  7. ы
  8. ящ
  9. edBy
  10. atic

Suppresses

  1. Roh
  2. =id
  3. eacher
  4. 特点是
  5. 他们是
  6. }\.[
  7. eña
  8. Superintendent
  9. =true
  10. PG