Neuron Explorer

Qwen3-0.6B

the spot before a number after 'less than' or 'more than'

A neuron in layer 5 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

The neuron fires on the spot just before a number in phrases like "less than", "more than", "over" or "about", where a quantity is about to follow. Turning it up produced text about numbers and counting, though it was not consistent.

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

    30 U.S. agencies involved themselves in war relief. The ARC, the oldest and most organized relief group in 1914, had less than 150 chapters and twenty thousand members. The agency had a new national headquarters

  2. #2 peak

    effusion. It is also done to relieve the shortness of breath that a large pleural effusion can cause.⏎Normally the pleural cavity contains less than 20 milliliters (4 teaspoons) of clear, yellowish (

  3. #3 peak

    tight squeeze through columns and a near-siphon to a low, broad breakdown room. A tunnel to the north drops 8 feet into a trench containing from 3 to 5 feet of water. Several side passages branch off from this section

  4. #4 peak

    Desert can be divided into three main geographic regions depending on their landscape: sand, plains and hills. There are also lakes and salt marshes. With less than 10 inches of rainfall in 2016, the temperature in this

  5. #5 peak

    single migratory event. If an earlier and ultimately failed voyage occurred, it left little trace in the genomes of people alive today.⏎The human genome contains more than 20,000 protein-coding genes, which carry the instructions for

  6. #6 peak

    in California, which represent more than 700,000 people. The Karuk is the second largest Native American tribe in California and has over 3,200 members. Their tribal lands include over 1.4

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: This, I have been thinking about the (1) in the morning. So, this is the, and so So I The So I So I So I So I So

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

    chakras all over our bodies; they are everywhere in and around us. They are all over the planet at sacred power points on the land. They are everywhere within and without.

  2. #2 peak

    two tribute sites Faces from the Lewis War Memorial and Second World War Tribute for Lewis.⏎The full list of war memorials in Lewis:<|endoftext|>I’ve seen my fair share of horrible PowerPoint presentations, and I’ve noticed that they tend to follow common

  3. #3 peak

    sstatement: said, “This is the inspiration for this year’s theme, Gender equality benefits everyone. It is a powerful reminder that we all have so much to gain when we advance gender equality. It is both the right thing to do and

  4. #4 peak

    once told Jacob (Gen. 26:4-5), “I will make your seed to multiply as the stars of heaven, and I will give to your seed all these countries. And in your seed shall all the nations of the

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. L3:1218 suppresses
  2. L3:293 weakly excites
  3. L4:2702 weakly excites
  4. L3:59 weakly excites
  5. L3:489 weakly suppresses
  6. L3:465 weakly suppresses
  7. L2:188 weakly excites
  8. L1:2006 weakly suppresses
  9. L0:3015 weakly excites
  10. L4:234 weakly excites
  11. L0:689 weakly suppresses
  12. L1:1152 weakly suppresses
  13. L3:753 weakly suppresses
  14. L0:2446 weakly suppresses
  15. L3:863 weakly suppresses
  16. L2:2607 weakly suppresses
  17. L0:311 weakly excites
  18. L4:684 weakly suppresses
  19. L2:302 weakly excites
  20. L4:1788 weakly excites
  21. L4:2562 weakly suppresses
  22. L3:2791 weakly excites
  23. L0:1187 weakly excites
  24. L3:1407 weakly suppresses

Feeds later layers

  1. L8:1029 excites
  2. L11:2346 excites
  3. L11:979 excites
  4. L20:1404 excites
  5. L10:141 suppresses
  6. L16:2448 excites
  7. L23:1404 excites
  8. L22:825 excites 'for' before a span of time
  9. L19:2851 excites
  10. L18:1006 weakly excites
  11. L10:960 weakly excites
  12. L17:2587 weakly excites
  13. L16:1154 weakly excites
  14. L17:2447 weakly excites
  15. L10:2300 weakly excites
  16. L19:701 weakly excites
  17. L16:2055 weakly excites
  18. L6:1697 weakly excites
  19. L19:1528 weakly excites
  20. L8:1368 weakly excites
  21. L12:2953 weakly suppresses
  22. L6:625 weakly excites
  23. L8:199 weakly suppresses
  24. L6:1582 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 -1.64 and -1.461 values between -1.28 and -1.101 values between -1.10 and -0.922 values between -0.92 and -0.7410 values between -0.74 and -0.5632 values between -0.56 and -0.38167 values between -0.38 and -0.201106 values between -0.20 and -0.022259 values between -0.02 and 0.16212 values between 0.16 and 0.3432 values between 0.34 and 0.5217 values between 0.52 and 0.7019 values between 0.70 and 0.8811 values between 0.88 and 1.069 values between 1.06 and 1.245 values between 1.24 and 1.423 values between 1.42 and 1.603 values between 1.60 and 1.782 values between 1.78 and 1.962 values between 1.96 and 2.142 values between 2.14 and 2.322 values between 2.32 and 2.501 values between 2.50 and 2.682 values between 2.68 and 2.861 values between 2.86 and 3.041 values between 3.22 and 3.401 values between 3.58 and 3.761 values between 3.94 and 4.131 values between 4.13 and 4.311 values between 5.39 and 5.57offnegativepositiveoften

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. оличество
  2. 만큼
  3. 足够的
  4. 足够
  5. imately
  6. enough
  7. ano
  8. anio
  9. aven
  10. casualties

Suppresses

  1. :border
  2. 说我
  3. 昧
  4. 我发现
  5. ::⏎
  6. ()?;⏎
  7. 还以为
  8. (environment
  9. 就知道
  10. 当我