Neuron Explorer

Qwen3-0.6B

"of" after quantity nouns (millions of, amount of)

A neuron in layer 20 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 "of" right after amount words such as "millions of", "amount of" or "heap of". Turned up, the model repeated "more" and "snow" rather than saying anything about amounts.

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

    , said Brian Jaskula, a lithium commodities specialist for the U.S. Geological Survey.⏎If successful, the operation could also generate millions of dollars to be used to combat the shrinking of the Salton Sea and the<|endoftext|>

  2. #2 peak

    than or equal to 1 mile, in just 1 year. Simply cutting out these short journeys could save millions of square metres of ice caps. How many of these trips could actually be made on

  3. #3 peak

    o file.o⏎Actually, the ``-o file.o'' is optional<|endoftext|>Photography is an approach of taking a photo with an electronic camera over an amount of time. An individual that produces pictures with a digital camera is called a digital photographer

  4. #4 peak

    lands offer nesting sites. Collisions between birds and aircraft, a.k.a. bird strikes, are a possible hazard. Not only do bird strikes lead to billions of euros in damages to aircraft, they

  5. #5 peak

    text. It is composed of a sequence of characters that define a search pattern. Regex can be used to find patterns in large amounts of text, validate user input, and manipulate strings. It is widely used in programming

  6. #6 peak

    web-based systems.⏎So, despite spending millions of dollars on these ERP systems, organizations are still being relegated to the use of manual

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 lot of more many more many more many... I have so many more many many many more many many much... etc. So I have so many more many many much, but I have so many more much, but 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

    diabetes and vascular disease placing prevention in the middle of the mental picture we would draw of a healthier world. Worldwide, the number of people with diabetes rose from 108 million in 1980 to 422 million in

  2. #2 peak

    Supplement, close to 20% of students deal with bullying problems. Similarly, the 2017 Youth Risk […]⏎Nurturing children is one of the challenging and rewarding responsibilities on earth— and you have to be prepared and

  3. #3 peak

    Lumbosacral Region of the SpineSkip to the navigation⏎The spine is composed of 33 interlocking bones called vertebrae. The lumbosacral region

  4. #4 peak

    very little work. Pop a handful of them in your mouth and chew the health benefits of almonds with joy.⏎Getting healthier by understanding nutrition is one of my main goals. I love eating almonds, especially because of how good they

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. L11:233 suppresses
  2. L17:1246 excites
  3. L18:2855 excites
  4. L13:1320 suppresses
  5. L6:2862 suppresses
  6. L4:2754 suppresses
  7. L6:1767 suppresses
  8. L8:1029 suppresses
  9. L18:2804 suppresses
  10. L12:1561 suppresses
  11. L10:167 suppresses
  12. L8:252 excites 'of' after 'each' or 'one' or 'all'
  13. L5:125 excites "of" after all three, both, ten
  14. L15:1554 excites
  15. L5:2980 excites 'host' or 'number' before 'of'
  16. L17:2447 suppresses
  17. L13:2114 suppresses
  18. L14:2846 suppresses
  19. L17:2348 excites
  20. L18:2183 suppresses
  21. L8:771 suppresses unclear: 'of' after a quantity noun
  22. L18:2608 suppresses
  23. L4:962 weakly excites
  24. L13:2715 weakly suppresses

Feeds later layers

  1. L22:496 excites
  2. L21:716 excites
  3. L21:274 suppresses
  4. L22:694 excites
  5. L24:770 excites
  6. L26:944 excites
  7. L24:703 excites
  8. L23:24 suppresses
  9. L21:696 excites
  10. L22:1077 suppresses
  11. L27:227 excites
  12. L27:477 excites
  13. L22:1228 excites
  14. L23:1967 weakly suppresses
  15. L21:68 weakly suppresses
  16. L22:2938 weakly suppresses
  17. L24:109 weakly suppresses
  18. L25:597 weakly suppresses
  19. L21:675 weakly excites
  20. L21:511 weakly excites
  21. L22:524 weakly suppresses
  22. L23:883 weakly suppresses
  23. L25:96 weakly excites
  24. L25:377 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 -10.77 and -9.941 values between -7.44 and -6.601 values between -6.60 and -5.771 values between -5.77 and -4.933 values between -4.93 and -4.103 values between -4.10 and -3.277 values between -3.27 and -2.4313 values between -2.43 and -1.60256 values between -1.60 and -0.763105 values between -0.76 and 0.07328 values between 0.07 and 0.9142 values between 0.91 and 1.7429 values between 1.74 and 2.5719 values between 2.57 and 3.4116 values between 3.41 and 4.247 values between 4.24 and 5.0810 values between 5.08 and 5.9110 values between 5.91 and 6.7410 values between 6.74 and 7.587 values between 7.58 and 8.415 values between 8.41 and 9.256 values between 9.25 and 10.085 values between 10.08 and 10.921 values between 10.92 and 11.755 values between 11.75 and 12.583 values between 12.58 and 13.426 values between 13.42 and 14.251 values between 14.25 and 15.092 values between 15.09 and 15.921 values between 16.75 and 17.592 values between 18.42 and 19.261 values between 19.26 and 20.091 values between 21.76 and 22.59offnegativepositiveoften

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. 话
  2. 人と
  3. people
  4. steps
  5. ยว
  6. 사람들이
  7. 年来
  8. MET
  9. instances
  10. 人が

Suppresses

  1. 臾
  2. enough
  3. edList
  4. bunch
  5. 了不少
  6. neath
  7. itur
  8. 多少
  9. plenty
  10. 已久的