Neuron Explorer

Qwen3-0.6B

'age' or 'way' ending a word

A neuron in layer 1 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 the last piece of a longer word when it is 'age' or 'way', as in 'passageways', 'headway' and 'drainages'. It seems to recognise word endings that are split off as a separate piece. Turning it up did not change the topic of what the model wrote.

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

    early and late stage asthma⏎Normally on inhalation of allergens, the airway narrows and wheezing develops within minutes as the lungs swell and passageways narrow. The wheezing will start in about 10 minutes,

  2. #2 peak

    people believe that the Pyramids are a link to an ancient lost civilization, and may still contain evidence to prove it. Many conspiracy theories regarding secret chambers and passageways within the pyramids exist to this day.⏎These conspiracy theories became even more

  3. #3 peak

    territory of 1057 km2. It shares its neighborhood with India’s World Heritage site, Manas Tiger Reserve, and is also associated by passageways to Phibsoo Wildlife Sanctuary and Jigme Singye National Park

  4. #4 peak

    |⏎|Day 4 – Spotting the Errors & Improving Sentences||3 hours||Essay Assessment|⏎|Day 5 – Grammar & Fixing Passages||3 hours||Complete Writing Exam|⏎How We Teach⏎For all aspects of

  5. #5 peak

    common as he traces the history of instant communication from ancient times to today.⏎→ Event: Tom Standage | Barnes and Noble Wednesday 6pm | Information⏎Excerpt: Writing on

  6. #6 peak

    on (just ask Joss Whedon or the Wachowski brothers or Dr. Octopus). So, it should be inevitable that companies start making headway towards bringing a revolutionary technology out of the hospitals and into the markets.⏎It’s

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 need to be able to think that the way I have been thinking about lately. So I need to be able to think that the way I have been thinking about lately. So I need to be able to think that the way I 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

    the primary problem lies with the heart muscle cells themselves; these diseases are called cardiomyopathies. Different cardiomyopathies are caused by different cellular defects, but most of the cardiomyopath

  2. #2 peak

    deficits (myosin V), and cell division defects that may be linked to cancer (nonmuscle myosin II). In addition, inherited familial cardiomyopathies affecting 1 out of 500 people result from missense

  3. #3 peak

    it.”⏎This is usually simplified as follows: small doses stimulate, medium doses inhibit and large doses kill. Thus the minimum, potentised dose is employed by homeopaths.⏎When seeing a homeopath for

  4. #4 peak

    - Individuals who have congestive heart failure of any cause⏎- Individuals with intrinsic heart muscle disease such as hypertrophic cardiomyopathy, idiopathic dilated cardiomyopathy, and arrhythmogenic right ventricular dysplasia⏎- Individuals with

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. L0:1301 suppresses
  2. L0:178 suppresses
  3. L0:1224 suppresses
  4. L0:2411 weakly excites
  5. L0:947 weakly suppresses
  6. L0:495 weakly suppresses
  7. L0:1066 weakly excites
  8. L0:1906 weakly suppresses
  9. L0:2090 weakly excites
  10. L0:1075 weakly excites
  11. L0:2232 weakly excites
  12. L0:1956 weakly suppresses
  13. L0:1355 weakly excites
  14. L0:71 weakly excites
  15. L0:1471 weakly excites
  16. L0:121 weakly excites
  17. L0:2770 weakly suppresses
  18. L0:2381 weakly excites
  19. L0:2651 weakly suppresses
  20. L0:940 weakly suppresses
  21. L0:2318 weakly suppresses
  22. L0:102 weakly suppresses
  23. L0:2601 weakly suppresses
  24. L0:2620 weakly suppresses

Feeds later layers

  1. L21:850 weakly excites
  2. L4:2551 weakly suppresses
  3. L7:2757 weakly excites
  4. L9:186 weakly excites
  5. L7:2439 weakly excites
  6. L11:93 weakly suppresses
  7. L24:1510 weakly excites
  8. L22:2899 weakly suppresses
  9. L11:484 weakly excites
  10. L15:1797 weakly excites
  11. L4:1532 weakly excites
  12. L17:2680 weakly excites
  13. L18:2673 weakly suppresses
  14. L19:1651 weakly suppresses
  15. L23:1646 weakly suppresses
  16. L20:41 weakly excites
  17. L20:1694 weakly excites
  18. L8:1445 weakly excites
  19. L8:1160 weakly suppresses
  20. L25:2776 weakly excites
  21. L14:678 weakly excites
  22. L10:2859 weakly suppresses
  23. L18:2011 weakly excites
  24. L12:990 weakly 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 -1.50 and -1.371 values between -1.11 and -0.991 values between -0.99 and -0.862 values between -0.73 and -0.602 values between -0.60 and -0.472 values between -0.47 and -0.3469 values between -0.34 and -0.21487 values between -0.21 and -0.082551 values between -0.08 and 0.05695 values between 0.05 and 0.1881 values between 0.18 and 0.308 values between 0.30 and 0.432 values between 0.43 and 0.561 values between 0.69 and 0.821 values between 0.82 and 0.952 values between 0.95 and 1.081 values between 2.24 and 2.361 values between 3.52 and 3.65offnegativepositiveoften

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. 银川
  2. mium
  3. ammu
  4. fred
  5. metic
  6. haled
  7. -vs
  8. brushing
  9. breaking
  10. acios

Suppresses

  1. ла
  2. Put
  3. Put
  4. ключения
  5. وار
  6. ела
  7. ハウス
  8. icers
  9. ror
  10. QP