Neuron Explorer

Qwen3-0.6B

'B' in vitamin names

A neuron in layer 7 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 letter 'B' when it follows the word 'vitamin', as in 'vitamin B12' or 'vitamin B1'. The strongest hits come from nutrition text. Its negative side appears on place-name fragments like 'Otago' and 'Copeland', which don't share the vitamin theme. Turning it up did not make the model talk about vitamins.

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

    his data during the 1769 transit of Venus, he was not exonerateed until a century later when the renowned American astronomer Simon Newcomb found Hell's readings to be correct, his scholarship above suspicion and his accusers

  2. #2 peak

    from the outer layer of the wheat grain. The cereal has a rich content of protein, fiber, vitamins, minerals and phytonutrients. Vitamins include B1, B2, B3, B6 and E. Mineral content includes

  3. #3 peak

    perfectly acceptable, but if you are planning to eat a purely vegetarian or vegan diet, the main nutrients of concern are protein, omega-3 fatty acids, vitamin B12,

  4. #4 peak

    .⏎Oats are one of the healthiest cereals on the planet. They are rich in healthy fiber, magnesium, manganese, thiamine (vitamin B1) and other important nutrients.⏎Protein content: 15% of

  5. #5 peak

    digestive track balanced. The nutritional composition of apples is very rich in vitamins, minerals, carbohydrates and amino acids etc. Apples are also very good source of vitamin B-Complex.⏎Apple contains a kind of antioxidant known as ‘quercetin

  6. #6 peak

    no animal source foods (ASFs). Vitamin B12 is found in red meat, seafood, dairy products and eggs, and

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 have a 200-year-old house in the middle of the village. It's a 19th-century cottage, built in 1791. It was a small house, with 400 square feet

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

    and having vastly different brain architectures — show such a skill as orthographic processing is astonishing.”⏎Another of the study’s authors, Professor Michael Colombo of Otago’s Department of Psychology, says “we may have to seriously re-think the

  2. #2 peak

    internal combustion engines.⏎The research will be conducted by the University's Powertrain & Vehicle Research Centre (PVRC) with both projects being led by Dr Colin Copeland in partnership with industry.⏎The first project, Additive Manufacturing for Cooled High

  3. #3 peak

    92 she passed the university junior scholarship examination.⏎In 1893 Ethel Benjamin enrolled for an LLB degree at the University of Otago. She was the first woman to be admitted to law school at the university and

  4. #4 peak

    with words⏎Pigeons can learn to distinguish real words from non-words by visually processing their letter combinations, surprising new research from the University of Otago in New Zealand and Ruhr University in Germany shows.⏎The researchers found

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. L3:198 strongly excites
  2. L4:1932 strongly suppresses
  3. L2:1488 strongly excites
  4. L3:223 strongly excites
  5. L2:1230 strongly excites
  6. L6:1748 suppresses
  7. L2:1820 suppresses
  8. L0:2458 excites
  9. L2:1979 suppresses
  10. L4:2588 excites
  11. L2:1142 suppresses
  12. L5:616 suppresses
  13. L3:2143 excites
  14. L3:1649 excites
  15. L6:1647 suppresses
  16. L6:2429 suppresses
  17. L6:593 excites
  18. L1:1129 excites
  19. L4:2664 excites
  20. L2:1734 excites
  21. L2:2613 suppresses
  22. L3:159 excites
  23. L2:1372 excites
  24. L6:2649 suppresses

Feeds later layers

  1. L17:664 strongly suppresses
  2. L10:664 strongly suppresses
  3. L9:1782 suppresses
  4. L23:2919 suppresses
  5. L16:903 suppresses
  6. L9:1386 suppresses
  7. L20:886 suppresses
  8. L22:1171 suppresses
  9. L21:1862 suppresses
  10. L22:527 suppresses
  11. L25:821 suppresses
  12. L17:1095 suppresses
  13. L25:1083 suppresses
  14. L25:550 suppresses
  15. L19:863 suppresses
  16. L8:1379 excites
  17. L8:1549 excites
  18. L8:514 suppresses
  19. L20:1637 suppresses
  20. L17:947 excites
  21. L26:1058 suppresses
  22. L26:1503 suppresses
  23. L19:634 excites
  24. L23:2451 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 -13.27 and -12.921 values between -7.28 and -6.921 values between -6.92 and -6.571 values between -5.51 and -5.161 values between -5.16 and -4.812 values between -4.81 and -4.451 values between -4.45 and -4.101 values between -4.10 and -3.752 values between -3.75 and -3.394 values between -3.04 and -2.691 values between -2.69 and -2.341 values between -2.34 and -1.986 values between -1.98 and -1.637 values between -1.63 and -1.2817 values between -1.28 and -0.9223 values between -0.92 and -0.5798 values between -0.57 and -0.222359 values between -0.22 and 0.131320 values between 0.13 and 0.4961 values between 0.49 and 0.84offnegativepositiveoften

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. INF
  2. Inf
  3. izando
  4. PositiveButton
  5. INIT
  6. INF
  7. 搏
  8. köp
  9. edriver
  10. intent

Suppresses

  1. lette
  2. lett
  3. Leicester
  4. velle
  5. £
  6. Wales
  7. combe
  8. cottage
  9. 伦敦
  10. Victorian