Neuron Explorer

Qwen3-0.6B

semicolons joining two clauses

A neuron in layer 4 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 semicolons that join two related sentences, especially in technical or translated-sounding text. It also dips on the 'ze' in 'Yangtze'. Turning it up or down did not produce a clear theme in the sample 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

    brushless DC motor is a single chip microcomputer control, commutation and commutation: position sensor output signal, through the software, control word control P1; Speed change: PWM control armature current; Current limiting control: voltage comparator.⏎

  2. #2 peak

    ountains are generally suitable for meeting places such as squares. It is an organic combination of music, water shape, and light to give people a visual and auditory beauty; at the same time, the

  3. #3 peak

    also integrated drive circuit.⏎Suitable for a certain speed but not high speed accuracy requirements of the field, such as fans, pumps, hair dryer and other applications; And most industrial control and automatic control applications, such as mixers, plasma processors

  4. #4 peak

    DC0 ~ 10v) interface and digital (serial transmission) interface. The analog interface is driven by a frequency converter and a three-phase asynchronous motor; the digital interface is controlled by a fully digital servo motor. Generally speaking, the

  5. #5 peak

    layer thickness, quality and uniformity; and it can be used to deposit a variety of different materials in thicknesses ranging

  6. #6 peak

    Jewish perspective, visiting Jewish historical sites in post-Holocaust, post-Communist Europe can be a very positive experience, emphasizing Jewish life, history and culture; but the experience also falls under what is now known as Dark Tourism—tourism

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 20th century. 1990. 1991. 1992. 1993. 1994. 1995. 1996

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

    , 2011) A study by seismologists at the China Seismological Bureau indicates that the massive Three Gorges dam on the Yangtze River caused a “significant” increase in seismic activity along the dam’s reservoir.⏎

  2. #2 peak

    smog. The affected cities, which include the capital Beijing, are located in the regions of Beijing-Tianjin-Hebei and of the northern Yangtze River delta - including the Jiangsu province - , as well as in the Fen

  3. #3 peak

    for Beijing.⏎(May 19, 2011) The world’s largest hydroelectric project was designed to tame the flood-prone Yangtze River and to generate clean energy

  4. #4 peak

    responsibility for the Third World’s debt crisis on the parties involved, instead of on First and Third World taxpayers. Pat also edited the English language translation of Yangtze! Yangtze!, the extraordinary critique by Chinese experts of the Three Gorges

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. L3:23 suppresses
  2. L3:672 excites
  3. L3:1216 weakly suppresses
  4. L3:130 weakly excites
  5. L3:294 weakly excites
  6. L3:1264 weakly excites
  7. L1:1547 weakly excites
  8. L2:609 weakly suppresses
  9. L3:249 weakly excites
  10. L3:2118 weakly excites
  11. L1:162 weakly excites
  12. L1:1594 weakly suppresses
  13. L3:50 weakly excites
  14. L3:691 weakly excites
  15. L1:2810 weakly suppresses
  16. L3:615 weakly suppresses
  17. L2:1899 weakly excites
  18. L2:377 weakly excites
  19. L1:699 weakly excites
  20. L3:113 weakly suppresses
  21. L0:776 weakly excites
  22. L0:1301 weakly suppresses
  23. L2:1613 weakly suppresses
  24. L2:2831 weakly excites

Feeds later layers

  1. L7:920 weakly suppresses
  2. L6:1121 weakly excites 'while' joining two clauses after a comma
  3. L18:69 weakly excites
  4. L19:2457 weakly excites
  5. L6:1630 weakly excites
  6. L15:2681 weakly suppresses
  7. L11:2003 weakly excites
  8. L6:973 weakly suppresses
  9. L12:1139 weakly excites
  10. L5:2626 weakly suppresses
  11. L5:315 weakly suppresses
  12. L7:2794 weakly suppresses
  13. L5:1584 weakly suppresses
  14. L5:939 weakly suppresses
  15. L6:255 weakly suppresses
  16. L27:845 weakly suppresses
  17. L9:1604 weakly suppresses
  18. L24:536 weakly suppresses
  19. L6:1817 weakly suppresses
  20. L25:2345 weakly suppresses
  21. L14:1614 weakly suppresses
  22. L7:2379 weakly suppresses
  23. L16:2463 weakly suppresses
  24. L12:1853 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 -1.31 and -1.231 values between -1.08 and -1.001 values between -1.00 and -0.921 values between -0.84 and -0.761 values between -0.68 and -0.611 values between -0.61 and -0.533 values between -0.53 and -0.454 values between -0.45 and -0.3734 values between -0.37 and -0.2999 values between -0.29 and -0.21266 values between -0.21 and -0.14644 values between -0.14 and -0.061126 values between -0.06 and 0.02968 values between 0.02 and 0.10453 values between 0.10 and 0.18180 values between 0.18 and 0.2682 values between 0.26 and 0.3428 values between 0.34 and 0.416 values between 0.41 and 0.493 values between 0.49 and 0.572 values between 0.57 and 0.651 values between 0.73 and 0.811 values between 0.96 and 1.041 values between 1.20 and 1.281 values between 1.75 and 1.82offnegativepositiveoften

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. ollipop
  2. engl
  3. олос
  4. uellement
  5. ivable
  6. 可观
  7. icrous
  8. lename
  9. kịch
  10. ueil

Suppresses

  1. ılmış
  2. IOUS
  3. d
  4. UNG
  5. stitute
  6. YS
  7. poverty
  8. IA
  9. specific
  10. INATION