Neuron Explorer

Qwen3-0.6B

blank spot right before a number

A neuron in layer 14 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 space just before a number or amount, such as the gap before '1.25mm', '70mm' or '1.4 billion', where the next word is likely to be a figure. Turning it up made the model's writing drift into repeated or odd number-heavy text, so the steering shows no clear theme.

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

    and Persuasion with Nothing But the Truth⏎|Grades||6 – 8|⏎|Lesson Plan Type||Standard Lesson|⏎|Estimated Time||Five 50-minute sessions|⏎MATERIALS AND TECHNOLOGY⏎Copies of Av

  2. #2 peak

    .25 x 1 Meter. The M8 (8mm) is the rod's diameter, the 1.25 means that there is a 1.25mm distance between each thread, and the 1 meter is

  3. #3 peak

    the Gemini IV mission during orbit no. 19, on June 4, 1965, with a Hasselblad camera and a 70mm lens. The Gemini IV crew — astronauts Jim McDivitt and

  4. #4 peak

    akes recommend getting from 45 to 65 percent of daily calories - from 900 to 1,300 calories in a typical 2,000-calorie diet

  5. #5 peak

    <|endoftext|>The New International Encyclopædia/Kinkajou⏎|←Kinic acid||The New International Encyclopædia⏎|Edition of 1905. See also Kinkajou on Wikipedia, and the

  6. #6 peak

    PN Bridge Rectifier Diode (x4)⏎- 1N5819 Schottky Diode⏎- Resistor (2MΩ, 560Ω, 1KΩ, 10Ω,

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: my own experience, my own experience, my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (This is my own experience. (

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

    017 WHSmith and its suppliers.⏎WHSmith High Street Limited Greenbridge Road, Swindon, Wiltshire, United Kingdom, SN3 3

  2. #2 peak

    an appointment with one of our tutors. If you have any questions, please contact us at firstname.lastname@example.org or 410.837.5383.<|endoftext|>Microsoft co-founder Paul Allen has shown off the “

  3. #3 peak

    155. 50-58. 10.1016/j.biocon.2012.05.013Full text not available from this repository.⏎Although grasslands are crucial habitats

  4. #4 peak

    Drive, Headington, Oxford, OX3 8QQ⏎Back to previous page<|endoftext|>A waterspout is a cyclonic wind

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. L11:226 suppresses
  2. L10:220 suppresses hyphen inside compound words
  3. L8:350 suppresses blank space before a year
  4. L4:53 excites blank slot before a number
  5. L12:1467 suppresses numbers after dates and figures
  6. L6:1469 excites
  7. L10:2461 excites
  8. L6:1563 excites
  9. L6:756 weakly suppresses
  10. L9:349 weakly suppresses the spot before a year
  11. L10:868 weakly suppresses
  12. L9:2160 weakly suppresses
  13. L10:2336 weakly excites
  14. L6:938 weakly excites
  15. L8:1244 weakly excites
  16. L6:679 weakly excites
  17. L4:209 weakly excites quantities right after 'a' in numbers
  18. L13:287 weakly suppresses
  19. L11:2532 weakly excites
  20. L9:2096 weakly suppresses 'the' or 'as' before a superlative role
  21. L13:1308 weakly excites
  22. L13:475 weakly suppresses
  23. L13:923 weakly suppresses word right before a number or year
  24. L6:1008 weakly excites 'of' after a measured quantity

Feeds later layers

  1. L15:368 excites
  2. L17:222 excites
  3. L16:256 excites number right after 'age of' or 'ages'
  4. L18:390 excites
  5. L19:58 suppresses
  6. L15:462 excites
  7. L16:211 excites
  8. L15:1909 excites blank slot before a number
  9. L17:46 excites
  10. L17:126 excites
  11. L15:448 excites
  12. L19:761 excites
  13. L16:253 weakly excites
  14. L21:557 weakly excites
  15. L26:154 weakly excites
  16. L21:175 weakly excites
  17. L16:372 weakly excites
  18. L17:124 weakly excites
  19. L15:828 weakly excites
  20. L22:9 weakly excites
  21. L15:1497 weakly excites
  22. L18:965 weakly excites
  23. L22:298 weakly excites
  24. L16:2137 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 -3.43 and -3.152 values between -1.78 and -1.505 values between -1.50 and -1.231 values between -1.23 and -0.955 values between -0.95 and -0.68130 values between -0.68 and -0.401665 values between -0.40 and -0.131793 values between -0.13 and 0.15168 values between 0.15 and 0.4253 values between 0.42 and 0.7023 values between 0.70 and 0.9712 values between 0.97 and 1.2510 values between 1.25 and 1.529 values between 1.52 and 1.804 values between 1.80 and 2.077 values between 2.07 and 2.353 values between 2.35 and 2.623 values between 2.62 and 2.902 values between 2.90 and 3.171 values between 3.17 and 3.452 values between 3.45 and 3.721 values between 3.72 and 4.001 values between 4.00 and 4.271 values between 4.55 and 4.823 values between 4.82 and 5.101 values between 5.65 and 5.921 values between 5.92 and 6.201 values between 7.30 and 7.57offnegativepositiveoften

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. ITO
  2. 多久
  3. LEGRO
  4. парти
  5. 一世
  6. 丹
  7. Builders
  8. ogan
  9. 苹果
  10. izo

Suppresses

  1. льн
  2. tram
  3. draulic
  4. รณ
  5. embedded
  6. vest
  7. Ń
  8. venes
  9. رب
  10. ncia