Neuron Explorer

Qwen3-0.6B

comma inside large numbers

A neuron in layer 19 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 lights up on the comma in big numbers like 36,000 or 90,000, where it marks the thousands. Turning it up made the model spit out long runs of commas, while turning it down made it write longer, looser sentences.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    expenses or deductions are subtracted out. Net income is always less than gross income.⏎For example, let’s say you get a job that pays $36,000 a year. You are paid twice a month, so each paycheck

  2. #2 peak

    0,000 square miles this year. The largest the area has ever been measured was in 2002, when it was about 8,500 square miles.⏎Scientists from the Louisiana Universities Marine Consortium said that this

  3. #3 peak

    ’s quality and usability. Regular updates and bug fixes based on user feedback contribute to long-term success. Lastly, keeping an eye on industry trends, exploring new technologies, and participating in developer communities foster growth and innovation

  4. #4 peak

    You are the performance management accountant, and you are requested to calculate the Sales Volume Variance of product A. The actual Sale for this product is 90,000 unit⏎Based on formula⏎- Sales volume variance = (Actual

  5. #5 peak

    uncertainty, serious risk, and even the potential for chaos. But there are also peaceful applications<|endoftext|>An estimated 50,000 to 60,000 farmworkers between the ages of 14 and 16

  6. #6 peak

    nonprofit The Nature Conservancy since the early 1980s to protect wildlife habitats. To date, Seattle City Light has purchased more than 10,000 acres to protect wildlife habitat, especially that of the various salmon and

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 but but the but but it and the the and the the and the and and then and then the and then and then and the and then and then the and then and then the and then and then the and then and then the and

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

    from China, snow peas from Guatemala and asparagus from Peru.⏎As fears of mad cow disease

  2. #2 peak

    on the screen here, are sturgeon down on the bottom of the river. Have you seen a st

  3. #3 peak

    screens versus books,” notes Wolf.⏎So screen-based reading is bad?⏎So is screen-based literacy “bad” and print-based literacy “good”? Not at all, says Wolf. Instead, she argues, “The great challenge now is to learn

  4. #4 peak

    University of Passau)<|endoftext|>||Audio Video Interleave File⏎||MPEG-4 Video File⏎||AVI is a video wrapper format created by Microsoft, storing video data that might be encoded with various codecs. AVI files can be

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. L16:240 strongly excites comma after 'sometimes' or 'the truth'
  2. L17:118 strongly excites comma after a clause ending
  3. L14:35 strongly suppresses
  4. L18:366 strongly excites comma inside numbers and dates
  5. L9:77 excites commas in citations and name lists
  6. L17:1405 suppresses
  7. L4:3 excites commas after citation fields
  8. L11:154 suppresses commas after citation numbers
  9. L17:116 excites
  10. L5:16 excites
  11. L10:335 excites commas in catalog, date, and list metadata
  12. L16:241 excites
  13. L18:1394 excites
  14. L16:1082 suppresses comma linking two clauses after 'and'
  15. L3:71 suppresses
  16. L15:588 excites comma after an introductory phrase
  17. L2:66 excites
  18. L18:892 suppresses
  19. L1:86 excites
  20. L16:489 excites
  21. L17:585 excites
  22. L18:48 suppresses
  23. L18:2376 suppresses
  24. L18:93 suppresses

Feeds later layers

  1. L23:184 strongly suppresses
  2. L22:144 suppresses
  3. L20:476 suppresses commas that lead into a continuation
  4. L21:632 excites
  5. L20:2082 excites
  6. L27:139 suppresses
  7. L26:94 suppresses
  8. L27:77 suppresses
  9. L21:700 suppresses comma before a new clause
  10. L22:368 excites
  11. L27:358 excites
  12. L26:236 excites
  13. L24:41 suppresses
  14. L20:2608 excites
  15. L27:191 excites
  16. L24:902 suppresses
  17. L21:50 suppresses comma after 'So' or intro phrase
  18. L26:22 suppresses the 'res' fragment inside words
  19. L20:811 excites
  20. L26:345 excites
  21. L27:130 suppresses
  22. L25:1221 excites
  23. L27:86 suppresses
  24. L25:118 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 -36.25 and -35.291 values between -34.34 and -33.382 values between -32.42 and -31.461 values between -31.46 and -30.501 values between -30.50 and -29.552 values between -29.55 and -28.596 values between -28.59 and -27.636 values between -27.63 and -26.674 values between -26.67 and -25.723 values between -25.72 and -24.765 values between -24.76 and -23.807 values between -23.80 and -22.846 values between -22.84 and -21.887 values between -21.88 and -20.939 values between -20.93 and -19.978 values between -19.97 and -19.016 values between -19.01 and -18.059 values between -18.05 and -17.105 values between -17.10 and -16.149 values between -16.14 and -15.185 values between -15.18 and -14.229 values between -14.22 and -13.2712 values between -13.27 and -12.3110 values between -12.31 and -11.357 values between -11.35 and -10.397 values between -10.39 and -9.434 values between -9.43 and -8.485 values between -8.48 and -7.526 values between -7.52 and -6.563 values between -6.56 and -5.606 values between -5.60 and -4.652 values between -4.65 and -3.694 values between -2.73 and -1.777 values between -1.77 and -0.82140 values between -0.82 and 0.143213 values between 0.14 and 1.10370 values between 1.10 and 2.06offnegativepositiveoften

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. ,
  2. ,...⏎⏎
  3. .—
  4. ,—
  5. ,b
  6. ,⏎⏎⏎⏎
  7. ,...⏎
  8. ,B
  9. ,W
  10. ,则

Suppresses

  1. regardless
  2. etc
  3. which
  4. especially
  5. but
  6. along
  7. including
  8. assuming
  9. pois
  10. 以及