Neuron Explorer

Qwen3-0.6B

commas in lists

A neuron in layer 15 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 commas and line breaks that separate items in a list, usually just before the last item or a conjunction such as 'and' or 'or'. Turning it up made the model produce long chains of 'or the mother, or the father', which fits the list idea.

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

    These include physical features of the<|endoftext|>Learn vocabulary, terms, and more with flashcards, games, and other study tools. Optional activities are designed to enhance understanding andor to provide

  2. #2 peak

    is one of these stories. In his novel, The Scarlet Letter, Nathaniel Hawthorne uses the symbolism of the Wild Rosebush, Hester’s Cabin, and the sunlight and the forest to contribute to the overall theme of imperfection.⏎

  3. #3 peak

    to sexual media content.⏎Source: University of Michigan⏎Explore further: New hadrosaur noses into spotlight<|endoftext|>bronzing, coating an object of wood, plaster, clay, or other substance to give it the colour and lustre of bronze.

  4. #4 peak

    actually modified stems, adept at storing water. They bloom during the spring in ranges of magentas and oranges, producing small to extra large magenta, yellow, and orange edible fruits (known in Mexico as Tunas) in the summertime

  5. #5 peak

    .⏎- Apply a killing dosage rate of Bioguard MSA<|endoftext|>OSMap: Hadrian's Wall, LR88.⏎Type: Wall Fort, Fort, Minor Settlement, Temple Or Shrine⏎|Wall: E (7) to Con

  6. #6 peak

    foreign relations--military and diplomatic information. Most of that information falls within five major areas: (1) military operations, (2) weapons technology, (3) diplomatic activities, (4) intelligence activities, and (5)

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 think it, I think it, I think I, I think I... maybe it's just, maybe it's still... I think I... maybe, I think I'm, I think... I think, maybe I might have,

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

    because English is presumed as the mainstream language of class instructions for accelerating<|endoftext|>People suffering from lactose intolerance are shifting to plant-based milk nowadays because plant-based milk like almond milk and soy milk doesn

  2. #2 peak

    in 2014, thirteen animals have tested positive for rabies in Erie County (five raccoons, five bats, and one each of a cat, skunk and fox).⏎On June 11th, a 1

  3. #3 peak

    ping babies.⏎In Kids Do, Animals Too ten pairs explain opposites to young children. Each pair features people and a different animal, including: fast, slow -- dogs in, out -- mice up, down

  4. #4 peak

    August 16, 2012) The antibacterial chemical triclosan, found in popular personal care products such as Colgate® Total

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. L14:1184 strongly suppresses
  2. L14:1689 suppresses
  3. L14:2298 suppresses
  4. L14:862 suppresses
  5. L14:1644 excites
  6. L14:1433 suppresses
  7. L14:2649 excites
  8. L14:613 excites
  9. L12:1454 excites
  10. L12:1651 excites
  11. L12:1835 suppresses
  12. L14:1354 excites
  13. L14:318 excites
  14. L14:1113 suppresses
  15. L14:1635 suppresses end of document token
  16. L10:122 excites
  17. L14:1382 excites
  18. L14:88 excites
  19. L14:464 weakly excites
  20. L14:485 weakly suppresses
  21. L12:690 weakly excites
  22. L14:1500 weakly suppresses
  23. L13:1562 weakly excites unclear: words that set up a qualifier
  24. L14:2341 weakly excites

Feeds later layers

  1. L25:1270 strongly excites
  2. L23:223 excites
  3. L20:1334 excites
  4. L24:205 suppresses
  5. L17:1897 excites
  6. L19:289 suppresses
  7. L16:357 excites
  8. L16:1541 suppresses
  9. L27:376 excites
  10. L25:41 suppresses
  11. L23:379 suppresses
  12. L17:1099 suppresses
  13. L16:2880 suppresses
  14. L26:779 excites
  15. L16:2992 excites
  16. L19:1325 excites
  17. L18:2362 excites
  18. L16:473 excites
  19. L16:774 suppresses
  20. L16:1388 suppresses
  21. L18:101 excites
  22. L16:3060 suppresses
  23. L27:289 suppresses
  24. L17:1856 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 -6.46 and -6.211 values between -5.47 and -5.231 values between -4.98 and -4.731 values between -4.73 and -4.481 values between -4.48 and -4.241 values between -3.99 and -3.744 values between -3.74 and -3.494 values between -3.49 and -3.253 values between -3.25 and -3.004 values between -3.00 and -2.759 values between -2.75 and -2.518 values between -2.51 and -2.2615 values between -2.26 and -2.0126 values between -2.01 and -1.7635 values between -1.76 and -1.5237 values between -1.52 and -1.2770 values between -1.27 and -1.0283 values between -1.02 and -0.78157 values between -0.78 and -0.53255 values between -0.53 and -0.28649 values between -0.28 and -0.031189 values between -0.03 and 0.21638 values between 0.21 and 0.46357 values between 0.46 and 0.71177 values between 0.71 and 0.9677 values between 0.96 and 1.2053 values between 1.20 and 1.4522 values between 1.45 and 1.707 values between 1.70 and 1.943 values between 1.94 and 2.198 values between 2.19 and 2.445 values between 2.44 and 2.693 values between 2.69 and 2.931 values between 2.93 and 3.183 values between 3.18 and 3.43offnegativepositiveoften

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. or
  2. 或
  3. 或是
  4. 或者
  5. oder
  6. 或
  7. 或者是
  8. 或其他
  9. eller
  10. 或者其他

Suppresses

  1. aland
  2. 和技术
  3. and
  4. &&
  5. 与时
  6. olta
  7. 并与
  8. &&
  9. 和发展
  10. ock