Neuron Explorer

Qwen3-0.6B

zeros in rounded thousands (e.g. 20,000)

A neuron in layer 17 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 zero digits inside large round numbers written with commas, such as the middle zeros in "23,000 subscribers". Turning it up gave writing that did not clearly relate to numbers, so the steering results do not confirm the pattern.

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

    and staying healthy, both physically and mentally.<|endoftext|>Tech moves fast! Stay ahead of the curve with Techopedia!⏎Join nearly 200,000 subscribers who receive actionable tech insights from Techopedia.⏎A vulnerability management system

  2. #2 peak

    that enhance public safety, and provides lifelong learning and technical exchange opportunities benefiting the global engineering and technology community. ASME has more than 127,000 members worldwide.⏎ASME′S VISION, MISSION STATEMENT AND

  3. #3 peak

    Crop Biotech Update is a weekly newsletter of ISAAA, a non-for-profit organization. The CBU is distributed for free to over 23,000 subscribers worldwide to inform them about the key developments in biosciences, especially

  4. #4 peak

    advanced students of cognitive development and epistemologists."<|endoftext|>Whangamarino Wetland is a lowland mosaic of wetland types covering almost 7,000 ha. It is the second

  5. #5 peak

    was very excited to discover the New Amigos board game. New Amigos makes language learning fun and interactive!⏎The game has sold over 42,000 copies in Norway where it was developed. In Europe, it is distributed through

  6. #6 peak

    : it will close at midnight on November 1st, shortly before the real election day on November 4. But already more than 47,000 people have voted, and the "result" is scarcely in doubt.⏎So

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: that the way in which I am * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

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 to stay active.⏎“As we age, our metabolism slows down anywhere from two to ten percent per decade,” Lewis continues. “It’s not like all of a sudden they start exercising that it’s going to do something. You have to have

  2. #2 peak

    exhibit unwanted behavior, such as excessive barking or destructive chewing and digging. Eliminating his natural behavior is unfair to your dog (not to mention almost impossible to accomplish), so you’ll find training easier if he has the

  3. #3 peak

    with her profound exploration of nature and the human experience. Her poems often evoke a sense of wonder and contemplation, inviting readers to reflect on their own lives. In this argumentative essay, we will delve into one of Mary Oliver’s poems,

  4. #4 peak

    or low-cost assistive technologies available online that work on any device, they won’t have to purchase expensive new equipment just to get started.⏎Thanks to technological advancement, people with disabilities or impairments can

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. L10:1052 suppresses
  2. L4:498 suppresses
  3. L13:53 excites period ending a claim before elaboration
  4. L2:1795 weakly suppresses
  5. L9:1911 weakly excites
  6. L15:115 weakly excites
  7. L16:177 weakly excites zero digit inside dates and codes
  8. L2:2044 weakly excites
  9. L0:2336 weakly suppresses
  10. L15:1012 weakly excites
  11. L16:507 weakly suppresses
  12. L16:35 weakly suppresses 'not' in 'but not limited to'
  13. L13:2379 weakly suppresses the digit zero in numbers
  14. L15:387 weakly excites
  15. L16:222 weakly suppresses
  16. L9:1712 weakly suppresses
  17. L10:273 weakly suppresses
  18. L15:937 weakly excites
  19. L5:1401 weakly excites
  20. L5:781 weakly suppresses
  21. L9:2447 weakly excites
  22. L15:89 weakly excites
  23. L10:320 weakly suppresses
  24. L6:75 weakly suppresses

Feeds later layers

  1. L24:1676 excites zeros in large round numbers
  2. L27:153 weakly suppresses the spot just before a historical number
  3. L25:473 weakly excites zeros inside large round numbers
  4. L21:1213 weakly excites
  5. L26:2339 weakly suppresses
  6. L19:479 weakly suppresses
  7. L25:1209 weakly suppresses
  8. L18:499 weakly suppresses
  9. L20:380 weakly suppresses
  10. L27:1396 weakly excites
  11. L24:376 weakly excites
  12. L21:1789 weakly suppresses
  13. L26:2154 weakly suppresses
  14. L19:1165 weakly suppresses
  15. L22:1503 weakly suppresses
  16. L26:3031 weakly suppresses
  17. L19:598 weakly suppresses
  18. L26:2223 weakly suppresses
  19. L18:1933 weakly suppresses
  20. L21:1298 weakly suppresses
  21. L18:105 weakly excites
  22. L23:1608 weakly suppresses
  23. L19:1790 weakly excites
  24. L26:2601 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.

2 values between -4.47 and -3.573 values between -2.67 and -1.7721 values between -1.77 and -0.872107 values between -0.87 and 0.031731 values between 0.03 and 0.9315 values between 0.93 and 1.824 values between 1.82 and 2.722 values between 2.72 and 3.621 values between 3.62 and 4.521 values between 4.52 and 5.421 values between 5.42 and 6.321 values between 6.32 and 7.213 values between 7.21 and 8.111 values between 8.11 and 9.013 values between 9.01 and 9.911 values between 9.91 and 10.811 values between 10.81 and 11.712 values between 11.71 and 12.601 values between 12.60 and 13.503 values between 14.40 and 15.301 values between 19.79 and 20.691 values between 24.28 and 25.181 values between 29.67 and 30.571 values between 30.57 and 31.47offnegativepositiveoften

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. Ethi
  2. ers
  3. 族自治县
  4. 普及
  5. red
  6. swiper
  7. 泛
  8. unded
  9. mid
  10. 炒股

Suppresses

  1. থ
  2. igua
  3. Boo
  4. isode
  5. patial
  6. acle
  7. theast
  8. рон
  9. esson
  10. ablish