Neuron Explorer

Qwen3-0.6B

word right before a number value

A neuron in layer 25 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 the word just before a number, especially ages, percentages, and measurements, such as "age" in "at age 67" or "between" in "between 30-45 percent". When it was turned up, the model's writing filled up with digits and numbers.

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

    organized the Wounded Knee uprising, a 71-day siege that included several gun battles with federal officers. Means died in 2012 at age 72.⏎“He was the only person in (a) leadership position in W

  2. #2 peak

    0, in addition to regular self-exams and exams by your doctor or other health care provider.⏎- Colon and rectal cancer: If you are over age 50, talk to your doctor about getting one or more of several screening tests

  3. #3 peak

    a young age).⏎Guidelines differ for men at average risk. Some recommend an initial PSA (Prostate Specific Antigen) blood test and clinical examination at age 40, and others recommend starting at age 50. In general,

  4. #4 peak

    have thought, “What is the ideal level for my house in the summer or winter months?”⏎During the summer months, the average humidity should weigh in between 30-45 percent (below the 50% mark). Winter

  5. #5 peak

    and the relative humidity need to be low.⏎Conventional air conditioning consists of setting the humidity at an acceptable level, while reducing the temperature. A humidity level of 50% is usually comfortable. The most common method of reducing humidity is to

  6. #6 peak

    Extremely diverse country (92 mother tongues) The majority of the population are Hindu Adult Literacy Rate (6 and over): 53% Life expectancy: 68 years Home of Mt. Everest and the highest altitude lake in the world

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: I think it is good service from a service company who values good service under good service tenasi (a tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi tenasi ten

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

    is unsurprising that women may experience an unplanned pregnancy or a pregnancy they cannot carry to term on more than one occasion. No form of contraception is 100 per cent effective, and women will always need straightforward access to abortion services as

  2. #2 peak

    on the face covering or use hand sanitiser with a minimum of 60% alcohol base.⏎3. Be sure the front of the face covering is

  3. #3 peak

    ! A candle made from soy wax burns cleaner, lasts longer and is less likely to trigger allergies! No pollutants, carcinogens or toxins! They are 100% biodegradable and on average burn longer than paraffin. (

  4. #4 peak

    free from the virus, especially during this COVID-19 outbreak.⏎- Regularly clean the device⏎- Use a disinfectant that contains at least 7

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. L23:775 strongly excites
  2. L21:1866 excites
  3. L21:1732 excites
  4. L23:2592 excites
  5. L23:2230 excites
  6. L23:2093 suppresses
  7. L22:2783 suppresses
  8. L24:2097 suppresses
  9. L22:2902 excites
  10. L23:2835 suppresses
  11. L22:1368 suppresses
  12. L21:2695 excites
  13. L22:2703 suppresses
  14. L24:992 excites
  15. L19:1851 excites
  16. L24:1527 suppresses space before a number, often an age
  17. L23:2791 suppresses
  18. L22:2854 excites
  19. L23:1858 weakly excites
  20. L16:1206 weakly excites
  21. L24:2785 weakly excites
  22. L22:1931 weakly suppresses
  23. L22:1574 weakly suppresses
  24. L21:1659 weakly excites

Feeds later layers

  1. L26:580 excites
  2. L27:205 suppresses
  3. L27:1196 suppresses
  4. L26:905 excites
  5. L26:598 suppresses spot just before a year or century
  6. L27:328 suppresses
  7. L26:754 weakly suppresses
  8. L27:530 weakly excites
  9. L27:1100 weakly excites
  10. L27:678 weakly excites
  11. L27:902 weakly suppresses
  12. L26:898 weakly excites
  13. L26:413 weakly suppresses
  14. L26:737 weakly suppresses
  15. L26:868 weakly excites
  16. L27:494 weakly suppresses
  17. L27:1153 weakly suppresses
  18. L26:1591 weakly suppresses
  19. L27:1897 weakly suppresses
  20. L27:180 weakly excites
  21. L26:1110 weakly excites
  22. L27:1232 weakly excites
  23. L26:1418 weakly excites
  24. L27:496 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 -31.80 and -29.931 values between -24.34 and -22.471 values between -22.47 and -20.611 values between -20.61 and -18.741 values between -16.88 and -15.013 values between -15.01 and -13.152 values between -11.28 and -9.412 values between -9.41 and -7.556 values between -5.68 and -3.8218 values between -3.82 and -1.951894 values between -1.95 and -0.091898 values between -0.09 and 1.7833 values between 1.78 and 3.6417 values between 3.64 and 5.5110 values between 5.51 and 7.377 values between 7.37 and 9.242 values between 9.24 and 11.102 values between 12.97 and 14.831 values between 14.83 and 16.701 values between 16.70 and 18.561 values between 18.56 and 20.431 values between 20.43 and 22.301 values between 24.16 and 26.032 values between 33.49 and 35.351 values between 35.35 and 37.221 values between 40.95 and 42.81offnegativepositiveoften

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. fort
  2. peria
  3. icens
  4. 丝绸
  5. 侗
  6. uming
  7. שת
  8. Researchers
  9. cestor
  10. 绸

Suppresses

  1. 🔟
  2. gether
  3. ランス
  4. way
  5. lay
  6. _ten
  7. Montserrat
  8. jure
  9. tenth
  10. DISCLAIMS