Neuron Explorer

Qwen3-0.6B

blank before or inside a number

A neuron in layer 23 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires at the spot just before a number, such as the gap in 'population of 1.3 billion', or on digits in years like 2011. It seems tied to places where a figure is about to appear. Turning it up made the model drift into repeated phrases and unrelated text.

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

    assigns electoral-college votes to each country (including the United States), based on the size of its population. So, for example, China with its population of 1.3 billion people gets 1,900 electoral-college votes

  2. #2 peak

    number of different Chinese “Dialects”, some which differ so much from each other that they may be referred to as “Languages”, for example Mandarin (about 850 million speakers) Cantonese (70 million). Other Major ones

  3. #3 peak

    overall economic growth.”⏎The Bureau of Labor Statistics says startups created about 4.65 million new jobs annually from 1997 to 2000 compared to 2.5 million in

  4. #4 peak

    demonstrated China's capability to conduct docking of a manned spacecraft with the experimental lab Tiangong 1 that China has had stationed in space since September 2011. Successful docking with an unmanned spacecraft had already been conducted in 20

  5. #5 peak

    approved the manned spaceflight programme.⏎The country has steadily moved to accomplish this and in fact, the white paper on space activities issued by China in 2011 categorically identified the national ambition to "launch space laboratories, manned spaceship and

  6. #6 peak

    over 3 million jobs being provided by the manufacturing industry, which employs 12 percent of the nation’s women and girls. That steady hum of progress

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 same thought, but with a different kind of construction, and with a different kind of construction, and with a different kind of construction, and with a different kind of construction. (I think that's a [building] [building]

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

    medicine⏎Knowing which genetic variants are responsible for differences in gene activity among individuals can give powerful clues for diagnosis, prognosis and intervention of different diseases. Senior author Emmanouil Dermitzakis, Louis Jeantet Professor at UNIGE

  2. #2 peak

    al. The script needed to be altered to meet these changes.⏎The Cham now live in two groups: the Western Cham of Cambodia and the Eastern Cham (Phan Rang Cham) of Vietnam. For

  3. #3 peak

    -entry of the unmanned Jules Verne ATV spaceship over the South Pacific.⏎"Jules Verne has now successfully completed its mission," ESA Director-General Jean-Jacques Dordain declared at the International Astronautical Congress in Glasgow, Scotland

  4. #4 peak

    the implications of non-cultural international legislation such as the European Convention on Human Rights and its fourteen protocols and national laws such as the Freedom of Information Act 2000;⏎- seek advice as necessary to ensure that the institution does what is right

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. L10:370 suppresses
  2. L4:267 weakly excites
  3. L4:42 weakly suppresses
  4. L18:233 weakly suppresses
  5. L7:92 weakly excites
  6. L2:24 weakly suppresses
  7. L4:1677 weakly excites
  8. L16:2154 weakly suppresses
  9. L17:1678 weakly excites
  10. L2:1853 weakly excites
  11. L22:2795 weakly suppresses
  12. L3:23 weakly suppresses
  13. L10:873 weakly excites
  14. L15:184 weakly excites
  15. L8:1222 weakly excites
  16. L3:1264 weakly excites
  17. L18:1307 weakly excites
  18. L8:328 weakly excites
  19. L11:43 weakly excites
  20. L17:943 weakly excites
  21. L22:1477 weakly excites
  22. L13:577 weakly suppresses
  23. L1:912 weakly excites
  24. L16:1044 weakly excites

Feeds later layers

  1. L27:1225 suppresses
  2. L27:2088 suppresses
  3. L25:618 suppresses
  4. L24:2855 suppresses
  5. L24:1559 suppresses
  6. L24:1114 weakly excites
  7. L25:2026 weakly suppresses
  8. L27:2721 weakly excites
  9. L24:418 weakly suppresses
  10. L24:1079 weakly excites
  11. L26:2700 weakly suppresses
  12. L24:1603 weakly excites
  13. L25:1718 weakly excites
  14. L24:1940 weakly excites
  15. L27:2810 weakly excites
  16. L25:1727 weakly suppresses
  17. L24:2426 weakly excites
  18. L25:2346 weakly excites
  19. L25:1083 weakly suppresses
  20. L25:1109 weakly excites
  21. L27:971 weakly excites
  22. L24:2294 weakly excites
  23. L25:2309 weakly suppresses
  24. L26:881 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.

1 values between -9.44 and -9.001 values between -5.94 and -5.501 values between -5.50 and -5.061 values between -4.63 and -4.191 values between -4.19 and -3.752 values between -3.31 and -2.886 values between -2.88 and -2.444 values between -2.44 and -2.005 values between -2.00 and -1.5613 values between -1.56 and -1.1380 values between -1.13 and -0.69502 values between -0.69 and -0.252233 values between -0.25 and 0.19901 values between 0.19 and 0.63127 values between 0.63 and 1.0613 values between 1.06 and 1.503 values between 1.50 and 1.944 values between 1.94 and 2.381 values between 2.38 and 2.811 values between 2.81 and 3.252 values between 3.25 and 3.691 values between 3.69 and 4.131 values between 4.56 and 5.001 values between 5.88 and 6.312 values between 7.19 and 7.631 values between 7.63 and 8.06offnegativepositiveoften

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. ointed
  2. ulos
  3. ewidth
  4. iode
  5. ulo
  6. ulary
  7. claimer
  8. thesis
  9. uchos
  10. ROL

Suppresses

  1. 建设
  2. 建設
  3. build
  4. construction
  5. 建设和
  6. building
  7. builds
  8. builder
  9. build
  10. Building