Neuron Explorer

Qwen3-0.6B

'a' or 'an' before a noun phrase

A neuron in layer 8 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 lights up on the small word 'a' or 'an' when it comes right after a preposition or verb, as in 'a variety of' or 'based on a'. When it was turned up, the text drifted into vague, generic wording, but the steered outputs did not clearly pick up the article theme.

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

    on the Atwater factors:⏎Fat: 9 cal/g Carb: 4 cal/g Protein: 4 cal/g⏎- Percent Daily Values are based on a 2,

  2. #2 peak

    ACS Nano.<|endoftext|>Regulating intestinal bacteria helps improve an individual’s stress status: that’s the conclusion reached recently by researchers from Shanghai University

  3. #3 peak

    herbal compounds, meditation, massage, and cleansing therapies. The people that practice Ayurvedic medicine claim that it prevents illnesses, treats serious health conditions and contributes to a healthy and long life.⏎The basic belief behind Ayurvedic medicine is that all

  4. #4 peak

    as Synodontis mossambica, Synodontis ornatus, Synodontis multipunctatus, and so on; some species are also farmed on a commercial scale as food or bait fish in some areas of Africa and Asia.⏎More

  5. #5 peak

    another over-the-counter medication to help combat nausea. Diphenhydramine (benadryl) is often helpful in treating nausea, motion sickness and a variety of skin issues. Dramamine or Bonine are also useful for dogs suffering

  6. #6 peak

    - Globally, 10-20 million people work in the ASGM sector and many of them use mercury on a daily basis.⏎- Countries from Africa, Asia-Pacific and the European Union too have

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: maybe I should have thought about something that is not just a regular math problem. So I decided to try something that are not regular math problems. So I decided to think about math problems that are not regular math problems. So I decided to

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

    for enjoyment as well as knowledge. The book started life as a series of five-minute radio broadcasts and for this reason its conversational style feels more like a series of fireside chats than a regular college textbook. Those radio broadcasts were highly acclaimed and

  2. #2 peak

    to Cape Lookout.<|endoftext|>Posted by Marcus on .⏎An ideal monatomic gas initially at 298 K and 1 atm pressure is⏎expanded from a volume of 1 liter to 2 liters. Consider separately⏎isobaric

  3. #3 peak

    911. It consists of about 200 buildings, which include dwellings and temples, a central plaza and a royal palace, all flanked by terraces for farming.⏎The stone for the buildings was mined from a quarry and

  4. #4 peak

    consequently took up arms.⏎At the Silveira farm near Punta Brava, Bandera made his last stand with twenty of his followers. He was surrounded by troops led by General Freyre Andrade, and a group of Rural Guards stormed

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. L6:2509 excites
  2. L6:640 excites "the" introducing a noun phrase after a name or preposition
  3. L6:1588 excites
  4. L4:137 excites
  5. L0:173 excites
  6. L7:385 suppresses
  7. L1:1062 suppresses
  8. L6:426 excites
  9. L4:1374 excites
  10. L5:457 suppresses 'a' after 'tell', 'bring' or 'much of'
  11. L5:249 weakly excites "a" right after "as"
  12. L5:1434 weakly excites
  13. L3:1810 weakly excites
  14. L7:7 weakly excites
  15. L4:2589 weakly suppresses
  16. L6:61 weakly suppresses
  17. L4:1573 weakly excites
  18. L3:787 weakly suppresses
  19. L6:1655 weakly suppresses "a" or "an" after "such" or launch verbs
  20. L7:121 weakly suppresses
  21. L4:326 weakly suppresses
  22. L6:806 weakly excites
  23. L4:2473 weakly excites
  24. L7:159 weakly suppresses

Feeds later layers

  1. L9:459 suppresses
  2. L14:313 excites
  3. L12:652 suppresses
  4. L10:578 suppresses
  5. L16:284 suppresses
  6. L9:59 suppresses "a" meaning per (a day, a year)
  7. L11:294 suppresses
  8. L14:1659 excites
  9. L24:150 suppresses
  10. L14:1402 suppresses
  11. L19:832 suppresses
  12. L13:975 weakly suppresses
  13. L19:183 weakly excites
  14. L18:2301 weakly suppresses
  15. L16:340 weakly suppresses
  16. L13:2325 weakly suppresses
  17. L12:72 weakly suppresses
  18. L9:330 weakly suppresses gap where a number was removed
  19. L13:1320 weakly suppresses
  20. L20:94 weakly suppresses
  21. L19:2119 weakly suppresses
  22. L17:935 weakly excites
  23. L11:2152 weakly excites
  24. L16:896 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 -4.22 and -4.001 values between -3.57 and -3.362 values between -2.50 and -2.294 values between -2.07 and -1.864 values between -1.86 and -1.645 values between -1.64 and -1.435 values between -1.43 and -1.2112 values between -1.21 and -1.009 values between -1.00 and -0.7820 values between -0.78 and -0.5762 values between -0.57 and -0.35406 values between -0.35 and -0.142352 values between -0.14 and 0.08836 values between 0.08 and 0.29144 values between 0.29 and 0.5128 values between 0.51 and 0.722 values between 0.72 and 0.944 values between 0.94 and 1.153 values between 1.15 and 1.373 values between 1.37 and 1.581 values between 1.80 and 2.011 values between 2.01 and 2.232 values between 2.23 and 2.441 values between 4.16 and 4.38offnegativepositiveoften

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. 做一个
  2. 印
  3. 果
  4. 作
  5. igua
  6. 体
  7. 报告
  8. ium
  9. 突
  10. 扰

Suppresses

  1. социально
  2. tsky
  3. .bundle
  4. lsen
  5. ensch
  6. ynchronously
  7. ationally
  8. zier
  9. 共产党员
  10. geführt