Neuron Explorer

Qwen3-0.6B

last digits of big numbers

A neuron in layer 7 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 reacts to the digits inside large numbers, like the final part of '24,300 mph' or '70,000 kilometres'. It does not change the model's writing in any obvious way when turned up.

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

    Telescope on the space rock as it zoomed away from Earth at 24,300 mph on Wednesday. Swift is best-known for its observations of high-energy outbur

  2. #2 peak

    Mars in several slingshot maneuvers to accelerate and help hurdle the spacecraft toward its destination. The comet is traveling toward the sun at 34,175 mph. After a high-speed chase across nearly 4 billion miles (6.

  3. #3 peak

    ospheric Multiscale mission (MMS) has set the Guinness World Record for highest altitude fix of a GPS signal – at 70,000 kilometres above the surface of the Earth.⏎Operating in a highly elliptical orbit around

  4. #4 peak

    Mars' distant past, according two new reports in the journal Nature.⏎Glaciers moved from the poles to the tropics 350,000 to 4 million years ago, depositing massive amounts of ice at the base

  5. #5 peak

    so-called transition zone (top) and the core-mantle boundary (below). Pressures in the lower mantle start at 237,000

  6. #6 peak

    above the convergent plate margin between the African Plate and the Eurasian Plate. It is the tallest active volcano in Europe, currently standing 3,329 m (10,922 ft) high, though this varies with

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: the first time, when we were in the first part of the story, we had to take into account for the future. What is the best way to have a place for them, and what is the best way for the future? -

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

    all counterfeit drugs contain the wrong ingredients, while 17 per cent contain the wrong levels of necessary ingredients.⏎High profits, low penalties⏎With nearly $1 trillion in<|endoftext|>|Scientific Name:||Pyrrhura viridicata

  2. #2 peak

    in Chicago was called upon to assemble some detailed cost figures from butter manufacturers whose 77 plants produced almost 150,000,000 pounds of butter annually. These figures reflected accurately the big increases in the costs of

  3. #3 peak

    05, over 7,000 natural disasters worldwide had taken the lives of nearly 2 million people and produced economic losses of more than $1 trillion. But the agency pointed out that efforts to better deal with disasters have begun to

  4. #4 peak

    forced to pay billions in bribes annually to numerous regulatory and police stops on its interstate highways.⏎Indian media has widely published allegations of corrupt Indian citizens stashing trillions of dollars in Swiss banks. Swiss authorities, however, deny these allegations.⏎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. L6:560 suppresses
  2. L5:2149 excites
  3. L5:866 suppresses
  4. L4:889 excites large number scale words
  5. L6:474 weakly suppresses
  6. L6:2059 weakly suppresses
  7. L1:439 weakly excites
  8. L6:2758 weakly suppresses
  9. L6:1946 weakly excites
  10. L3:2813 weakly excites
  11. L6:2559 weakly suppresses
  12. L6:2227 weakly excites
  13. L4:2922 weakly suppresses
  14. L6:1326 weakly suppresses
  15. L6:1878 weakly excites
  16. L6:1867 weakly suppresses
  17. L6:1119 weakly suppresses
  18. L4:2376 weakly suppresses
  19. L5:723 weakly excites
  20. L6:2160 weakly suppresses
  21. L6:1193 weakly excites
  22. L1:2719 weakly excites
  23. L0:1223 weakly excites
  24. L3:2323 weakly suppresses

Feeds later layers

  1. L11:2125 excites
  2. L10:589 suppresses
  3. L12:1194 suppresses
  4. L12:1867 weakly excites
  5. L9:2753 weakly suppresses
  6. L21:882 weakly excites
  7. L25:2795 weakly suppresses
  8. L23:2327 weakly suppresses
  9. L13:2537 weakly excites
  10. L15:2548 weakly excites
  11. L23:442 weakly suppresses
  12. L18:2269 weakly suppresses
  13. L15:1039 weakly excites
  14. L10:725 weakly excites
  15. L23:2085 weakly excites
  16. L10:650 weakly excites
  17. L12:843 weakly suppresses
  18. L10:1559 weakly excites
  19. L13:730 weakly suppresses
  20. L11:1040 weakly excites
  21. L10:2550 weakly excites
  22. L23:2136 weakly excites 'in' after a noun phrase
  23. L13:2629 weakly suppresses
  24. L25:2055 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 -3.27 and -3.111 values between -2.15 and -1.992 values between -1.67 and -1.511 values between -1.51 and -1.364 values between -1.04 and -0.888 values between -0.88 and -0.728 values between -0.72 and -0.5619 values between -0.56 and -0.40134 values between -0.40 and -0.24826 values between -0.24 and -0.082073 values between -0.08 and 0.08644 values between 0.08 and 0.24122 values between 0.24 and 0.4034 values between 0.40 and 0.568 values between 0.56 and 0.725 values between 0.72 and 0.884 values between 0.88 and 1.043 values between 1.04 and 1.203 values between 1.20 and 1.363 values between 1.36 and 1.522 values between 1.68 and 1.841 values between 1.84 and 2.001 values between 2.79 and 2.951 values between 2.95 and 3.11offnegativepositiveoften

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. onymous
  2. ctype
  3. bote
  4. kiện
  5. ッシュ
  6. villa
  7. chet
  8. atomy
  9. eldo
  10. stown

Suppresses

  1. ither
  2. Hers
  3. 出让
  4. atar
  5. licensors
  6. рав
  7. entert
  8. landır
  9. alore
  10. iard