Neuron Explorer

Qwen3-0.6B

blank slot before a number

A neuron in layer 15 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 spot just before a number or percentage, such as "affects about [] 25%". The model is effectively waiting for a figure to come next. Turned up, the model wrote lots of numbers and years, which fits this.

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

    Treatment may require hospitalization.⏎HG is a more severe form of nausea and vomiting of pregnancy (NVP), also called morning sickness. Morning sickness affects anywhere between 50% to 90% of pregnant women. HG is estimated to

  2. #2 peak

    it is diagnosed early. And that is why a prostate checkup is so imperative.⏎Survival rates for prostate cancer diagnosed and managed in its earliest stages are > 80 %. Regretfully, because symptoms are often non-existent at the early

  3. #3 peak

    , however, many people recover once they’re removed from a contaminated environment.⏎Chronic illness only occurs in those who are genetically susceptible to improper toxin clearance (about 25% of the population). This minority of people typically have genes that prevent

  4. #4 peak

    lanfair D.C., Llanelidan and Derwen,<|endoftext|>Ulcers are a common scourge in horses. Gastric ulcers affect up to 93 percent of racehorses, about 60 percent of horses in

  5. #5 peak

    vomiting blood or passing out altered blood in the stools. Of course, there is the dreaded complication of liver cancer.⏎Generally, cirrhosis appears in at least 20% of patients within 20 years of infection. Cir<|endoftext|>Parts

  6. #6 peak

    pain, especially when treatment is delayed. In fact, the type of spinal curvature called scoliosis is a relatively common cause of chronic back pain, affecting about 2-3% of Americans.⏎While

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: 100 years. I have 100 years of 2020 to 2020 years, 100 years to 2020 years. I have 2020 years

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

    checking if the input sequence is sorted and if not swapping randomly two elements. This is repeated until eventually the sequence is sorted.⏎If we consider all permutations of the first 4 natural numbers as input the expectation value of the number of swaps, averaged

  2. #2 peak

    Tom Standage, digital editor at the Economist, author of A History of the World in 6 Glasses and Writing on the Wall: Social Media - The First 2,000 Years (Bloomsbury USA, 201

  3. #3 peak

    lender may require additional documentation or information to assess the borrower's financial stability. In this case, the lender may use an average of the borrower's income over the past few years or take into account the borrower's other sources of income to determine their

  4. #4 peak

    today is a paradigm: the viewpoint of the defeated.⏎The book published in 1959 is part of the University Student Library, and to celebrate its 50 years the National Council for Culture and

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. L14:2494 strongly suppresses
  2. L10:868 suppresses
  3. L14:1898 excites
  4. L6:1469 excites
  5. L13:1308 excites
  6. L14:2425 suppresses
  7. L14:2643 excites
  8. L14:471 excites
  9. L14:1911 suppresses
  10. L12:2110 suppresses
  11. L6:756 suppresses
  12. L6:1411 suppresses
  13. L11:2709 suppresses
  14. L13:2125 suppresses
  15. L13:1628 excites
  16. L14:2757 excites 'by' giving the amount of a difference
  17. L14:840 suppresses
  18. L12:2797 excites
  19. L10:2461 excites
  20. L14:1435 excites
  21. L13:513 excites
  22. L9:1108 suppresses
  23. L13:2915 suppresses
  24. L13:1083 excites

Feeds later layers

  1. L17:1570 suppresses
  2. L21:175 suppresses
  3. L17:222 suppresses
  4. L16:394 suppresses
  5. L19:1419 suppresses
  6. L18:291 excites
  7. L17:1899 suppresses
  8. L25:2902 suppresses
  9. L21:676 suppresses
  10. L19:1222 suppresses
  11. L16:1897 suppresses
  12. L25:46 excites
  13. L16:372 suppresses
  14. L21:1687 suppresses
  15. L18:390 suppresses
  16. L19:1947 suppresses
  17. L20:945 suppresses
  18. L26:667 suppresses
  19. L16:2059 suppresses
  20. L16:1169 suppresses
  21. L19:2151 suppresses
  22. L17:932 suppresses
  23. L20:1682 suppresses
  24. L26:172 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 -4.36 and -4.081 values between -2.97 and -2.701 values between -1.59 and -1.313 values between -1.31 and -1.043 values between -1.04 and -0.7632 values between -0.76 and -0.48443 values between -0.48 and -0.202350 values between -0.20 and 0.07933 values between 0.07 and 0.3584 values between 0.35 and 0.6317 values between 0.63 and 0.918 values between 0.91 and 1.188 values between 1.18 and 1.465 values between 1.46 and 1.743 values between 1.74 and 2.014 values between 2.01 and 2.291 values between 2.57 and 2.841 values between 2.84 and 3.122 values between 3.12 and 3.402 values between 3.68 and 3.951 values between 3.95 and 4.231 values between 4.23 and 4.511 values between 4.51 and 4.781 values between 4.78 and 5.061 values between 6.45 and 6.72offnegativepositiveoften

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. W
  2. rush
  3. 簸
  4. most
  5. w
  6. liest
  7. 羡
  8. 热
  9. pady
  10. sendo

Suppresses

  1. .toArray
  2. seventy
  3. 五六
  4. 十条
  5. 一百
  6. ntax
  7. ninety
  8. 五百
  9. 二百
  10. 七八