Neuron Explorer

Qwen3-0.6B

hyphen in number or size compounds

A neuron in layer 26 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

It fires on a hyphen that joins a number or measurement to a word, like '1-1 lesson', '8-lead package', '100-gram serving' or 'full-fledged'. It goes quiet, or dips below zero, on hyphens in long chains of words like 'one-at-a-time' or 'quality-of-life'. Turning it up made the model repeat hyphenated word pairs until the text broke down.

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

    taught that you cannot compress liquids, so the the short answer to your question is yes, liquids have a definite volume :0).⏎Interested in booking a 1-1 lesson with me?Click here to view my profile and send me a message

  2. #2 peak

    V DAC. Total board area for the boost regulator circuit is only 43 mm2. The ADP1613 is available in an 8-lead MSOP package and the AD5504 in a 16

  3. #3 peak

    36 inches. Velocity slows to 1,071 and energy drops to 636 lb/ft.⏎This is a 45-70 Ballistics chart for rifles with a rifle<|endoftext|>Recently, the Union Ministry

  4. #4 peak

    , zinc, flavonoids, B vitamins, and fiber. It is also low in calories.⏎Nutritional Value and Health Effects⏎A 100-gram (3.5 oz) serving of oca provides almost 40

  5. #5 peak

    the "Navajo nation" is not a non-nation nation such as proposed for Quebec, but is a quasi-independent nation. Navajos are also full-fledged American citizens as they should be since their ancestors predate those of most others

  6. #6 peak

    in war. The U.S. military was also developing miniature UAVs that could be launched and operated by one person. The Marine Corps began testing a 2-kg (4.4-l

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: he might need to-scale-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time-time

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

    two-dimensional materials, by introducing individual defects in the material structure. By irradiating a defect with a laser pulse, single photons are emitted from the defect one-at-a-time and on demand. This project seeks to demonstrate that it is also possible to

  2. #2 peak

    . After adjusting for age, gender, hypertension, diabetes and obstructive sleep apnea, sleep quality and daytime sleepiness were significantly associated with mood disturbance and quality-of-life impairment.⏎“Our findings suggest that routine screening for sleep disorders in

  3. #3 peak

    lass urban canyons of cities can deliver cleaner air at the roadside, where most of us are exposed to the highest pollution levels, and can be implemented street-by-street without the need for large-scale and expensive initiatives.⏎So Transport for London

  4. #4 peak

    andibular joint is located on either side of your head in front of your ears. This joint allows you to move your jaw up and down and from side-to-side. With TMJD, it becomes difficult

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. L25:330 strongly suppresses
  2. L23:1175 strongly suppresses
  3. L23:1652 strongly suppresses
  4. L20:1064 strongly suppresses
  5. L22:494 strongly excites
  6. L20:2666 strongly excites
  7. L20:1571 suppresses
  8. L25:149 suppresses
  9. L19:767 suppresses
  10. L24:733 excites
  11. L19:2267 suppresses hyphen joining compound words
  12. L23:3002 suppresses
  13. L21:586 excites
  14. L19:1053 excites
  15. L23:2575 suppresses
  16. L25:132 excites
  17. L19:2301 suppresses
  18. L19:2307 excites
  19. L24:2590 excites
  20. L16:899 suppresses
  21. L22:2535 excites
  22. L23:19 excites
  23. L24:95 excites
  24. L19:430 excites

Feeds later layers

  1. L27:106 excites
  2. L27:1098 excites
  3. L27:166 excites
  4. L27:1459 suppresses
  5. L27:334 weakly excites
  6. L27:213 weakly excites
  7. L27:816 weakly excites
  8. L27:161 weakly suppresses
  9. L27:1432 weakly suppresses
  10. L27:362 weakly excites
  11. L27:662 weakly excites
  12. L27:1992 weakly excites
  13. L27:1206 weakly excites
  14. L27:1113 weakly suppresses
  15. L27:1556 weakly excites
  16. L27:1554 weakly suppresses
  17. L27:1778 weakly excites
  18. L27:2930 weakly suppresses
  19. L27:431 weakly excites
  20. L27:1007 weakly suppresses
  21. L27:595 weakly excites
  22. L27:669 weakly excites
  23. L27:82 weakly excites
  24. L27:726 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.

2 values between -61.06 and -58.831 values between -56.59 and -54.351 values between -27.51 and -25.271 values between -20.80 and -18.565 values between -16.32 and -14.084 values between -11.85 and -9.615 values between -9.61 and -7.3711 values between -7.37 and -5.1418 values between -5.14 and -2.90241 values between -2.90 and -0.663517 values between -0.66 and 1.5857 values between 1.58 and 3.8115 values between 3.81 and 6.0516 values between 6.05 and 8.292 values between 8.29 and 10.535 values between 10.53 and 12.761 values between 12.76 and 15.003 values between 15.00 and 17.241 values between 17.24 and 19.471 values between 19.47 and 21.711 values between 26.19 and 28.42offnegativepositiveoften

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. _unix
  2. _Local
  3. .Normal
  4. _wh
  5. [Index
  6. (p
  7. (b
  8. those
  9. conductor
  10. -A

Suppresses

  1. -level
  2. -scale
  3. -grade
  4. -style
  5. -time
  6. -shaped
  7. -side
  8. -quality
  9. -sized
  10. -quarters